diff options
44 files changed, 2211 insertions, 373 deletions
diff --git a/.gitignore b/.gitignore index f6c3927a2..2b6b5281a 100644 --- a/.gitignore +++ b/.gitignore @@ -322,4 +322,6 @@ build.sh !core/debug/ # RAD debugger project file -*.raddbg
\ No newline at end of file +*.raddbg + +misc/featuregen/featuregen diff --git a/base/intrinsics/intrinsics.odin b/base/intrinsics/intrinsics.odin index dca33bfd9..0c4e5d6c3 100644 --- a/base/intrinsics/intrinsics.odin +++ b/base/intrinsics/intrinsics.odin @@ -169,15 +169,18 @@ type_has_nil :: proc($T: typeid) -> bool --- type_is_specialization_of :: proc($T, $S: typeid) -> bool --- -type_is_variant_of :: proc($U, $V: typeid) -> bool where type_is_union(U) --- -type_union_tag_type :: proc($T: typeid) -> typeid where type_is_union(T) --- -type_union_tag_offset :: proc($T: typeid) -> uintptr where type_is_union(T) --- -type_union_base_tag_value :: proc($T: typeid) -> int where type_is_union(U) --- -type_union_variant_count :: proc($T: typeid) -> int where type_is_union(T) --- -type_variant_type_of :: proc($T: typeid, $index: int) -> typeid where type_is_union(T) --- -type_variant_index_of :: proc($U, $V: typeid) -> int where type_is_union(U) --- - -type_has_field :: proc($T: typeid, $name: string) -> bool --- +type_is_variant_of :: proc($U, $V: typeid) -> bool where type_is_union(U) --- +type_union_tag_type :: proc($T: typeid) -> typeid where type_is_union(T) --- +type_union_tag_offset :: proc($T: typeid) -> uintptr where type_is_union(T) --- +type_union_base_tag_value :: proc($T: typeid) -> int where type_is_union(U) --- +type_union_variant_count :: proc($T: typeid) -> int where type_is_union(T) --- +type_variant_type_of :: proc($T: typeid, $index: int) -> typeid where type_is_union(T) --- +type_variant_index_of :: proc($U, $V: typeid) -> int where type_is_union(U) --- + +type_bit_set_elem_type :: proc($T: typeid) -> typeid where type_is_bit_set(T) --- +type_bit_set_underlying_type :: proc($T: typeid) -> typeid where type_is_bit_set(T) --- + +type_has_field :: proc($T: typeid, $name: string) -> bool --- type_field_type :: proc($T: typeid, $name: string) -> typeid --- type_proc_parameter_count :: proc($T: typeid) -> int where type_is_proc(T) --- @@ -282,6 +285,12 @@ simd_reverse :: proc(a: #simd[N]T) -> #simd[N]T --- simd_rotate_left :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T --- simd_rotate_right :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T --- +// Checks if the current target supports the given target features. +// +// Takes a constant comma-seperated string (eg: "sha512,sse4.1"), or a procedure type which has either +// `@(require_target_feature)` or `@(enable_target_feature)` as its input and returns a boolean indicating +// if all listed features are supported. +has_target_feature :: proc($test: $T) -> bool where type_is_string(T) || type_is_proc(T) --- // WASM targets only wasm_memory_grow :: proc(index, delta: uintptr) -> int --- @@ -293,9 +302,9 @@ wasm_memory_size :: proc(index: uintptr) -> int --- // 0 - indicates that the thread blocked and then was woken up // 1 - the loaded value from `ptr` did not match `expected`, the thread did not block // 2 - the thread blocked, but the timeout -@(enable_target_feature="atomics") +@(require_target_feature="atomics") wasm_memory_atomic_wait32 :: proc(ptr: ^u32, expected: u32, timeout_ns: i64) -> u32 --- -@(enable_target_feature="atomics") +@(require_target_feature="atomics") wasm_memory_atomic_notify32 :: proc(ptr: ^u32, waiters: u32) -> (waiters_woken_up: u32) --- // x86 Targets (i386, amd64) diff --git a/build_odin.sh b/build_odin.sh index cacd1a7ed..7e0d0e585 100755 --- a/build_odin.sh +++ b/build_odin.sh @@ -2,7 +2,6 @@ set -eu : ${CPPFLAGS=} -: ${CXX=clang++} : ${CXXFLAGS=} : ${LDFLAGS=} : ${LLVM_CONFIG=} @@ -26,12 +25,14 @@ error() { if [ -z "$LLVM_CONFIG" ]; then # darwin, linux, openbsd - if [ -n "$(command -v llvm-config-17)" ]; then LLVM_CONFIG="llvm-config-17" + if [ -n "$(command -v llvm-config-18)" ]; then LLVM_CONFIG="llvm-config-18" + elif [ -n "$(command -v llvm-config-17)" ]; then LLVM_CONFIG="llvm-config-17" elif [ -n "$(command -v llvm-config-14)" ]; then LLVM_CONFIG="llvm-config-14" elif [ -n "$(command -v llvm-config-13)" ]; then LLVM_CONFIG="llvm-config-13" elif [ -n "$(command -v llvm-config-12)" ]; then LLVM_CONFIG="llvm-config-12" elif [ -n "$(command -v llvm-config-11)" ]; then LLVM_CONFIG="llvm-config-11" # freebsd + elif [ -n "$(command -v llvm-config18)" ]; then LLVM_CONFIG="llvm-config18" elif [ -n "$(command -v llvm-config17)" ]; then LLVM_CONFIG="llvm-config17" elif [ -n "$(command -v llvm-config14)" ]; then LLVM_CONFIG="llvm-config14" elif [ -n "$(command -v llvm-config13)" ]; then LLVM_CONFIG="llvm-config13" @@ -44,21 +45,22 @@ if [ -z "$LLVM_CONFIG" ]; then fi fi +: ${CXX=$($LLVM_CONFIG --bindir)/clang++} + LLVM_VERSION="$($LLVM_CONFIG --version)" LLVM_VERSION_MAJOR="$(echo $LLVM_VERSION | awk -F. '{print $1}')" LLVM_VERSION_MINOR="$(echo $LLVM_VERSION | awk -F. '{print $2}')" LLVM_VERSION_PATCH="$(echo $LLVM_VERSION | awk -F. '{print $3}')" -if [ $LLVM_VERSION_MAJOR -lt 11 ] || - ([ $LLVM_VERSION_MAJOR -gt 14 ] && [ $LLVM_VERSION_MAJOR -lt 17 ]); then - error "Invalid LLVM version $LLVM_VERSION: must be 11, 12, 13, 14 or 17" +if [ $LLVM_VERSION_MAJOR -lt 11 ] || ([ $LLVM_VERSION_MAJOR -gt 14 ] && [ $LLVM_VERSION_MAJOR -lt 17 ]) || [ $LLVM_VERSION_MAJOR -gt 18 ]; then + error "Invalid LLVM version $LLVM_VERSION: must be 11, 12, 13, 14, 17 or 18" fi case "$OS_NAME" in Darwin) if [ "$OS_ARCH" = "arm64" ]; then - if [ $LLVM_VERSION_MAJOR -lt 13 ] || [ $LLVM_VERSION_MAJOR -gt 17 ]; then - error "Darwin Arm64 requires LLVM 13, 14 or 17" + if [ $LLVM_VERSION_MAJOR -lt 13 ]; then + error "Invalid LLVM version $LLVM_VERSION: Darwin Arm64 requires LLVM 13, 14, 17 or 18" fi fi diff --git a/core/math/big/helpers.odin b/core/math/big/helpers.odin index 35be4f1fd..1969fac49 100644 --- a/core/math/big/helpers.odin +++ b/core/math/big/helpers.odin @@ -356,7 +356,7 @@ int_count_lsb :: proc(a: ^Int, allocator := context.allocator) -> (count: int, e } platform_count_lsb :: #force_inline proc(a: $T) -> (count: int) - where intrinsics.type_is_integer(T) && intrinsics.type_is_unsigned(T) { + where intrinsics.type_is_integer(T), intrinsics.type_is_unsigned(T) { return int(intrinsics.count_trailing_zeros(a)) if a > 0 else 0 } diff --git a/core/math/big/internal.odin b/core/math/big/internal.odin index 07e1e6c03..fa747e46a 100644 --- a/core/math/big/internal.odin +++ b/core/math/big/internal.odin @@ -546,7 +546,7 @@ internal_int_shl1 :: proc(dest, src: ^Int, allocator := context.allocator) -> (e Like `internal_int_mul_digit` but with an integer as the small input. */ internal_int_mul_integer :: proc(dest, a: ^Int, b: $T, allocator := context.allocator) -> (err: Error) -where intrinsics.type_is_integer(T) && T != DIGIT { +where intrinsics.type_is_integer(T), T != DIGIT { context.allocator = allocator t := &Int{} @@ -2806,7 +2806,7 @@ internal_int_count_lsb :: proc(a: ^Int) -> (count: int, err: Error) { } internal_platform_count_lsb :: #force_inline proc(a: $T) -> (count: int) - where intrinsics.type_is_integer(T) && intrinsics.type_is_unsigned(T) { + where intrinsics.type_is_integer(T), intrinsics.type_is_unsigned(T) { return int(intrinsics.count_trailing_zeros(a)) if a > 0 else 0 } diff --git a/core/math/big/radix.odin b/core/math/big/radix.odin index 8d8ea734e..f4eed879f 100644 --- a/core/math/big/radix.odin +++ b/core/math/big/radix.odin @@ -469,7 +469,7 @@ internal_int_pack_count :: proc(a: ^Int, $T: typeid, nails := 0) -> (size_needed Assumes `a` not to be `nil` and to have been initialized. */ internal_int_pack :: proc(a: ^Int, buf: []$T, nails := 0, order := Order.LSB_First) -> (written: int, err: Error) - where intrinsics.type_is_integer(T) && intrinsics.type_is_unsigned(T) && size_of(T) <= 16 { + where intrinsics.type_is_integer(T), intrinsics.type_is_unsigned(T), size_of(T) <= 16 { assert(nails >= 0 && nails < (size_of(T) * 8)) @@ -505,7 +505,7 @@ internal_int_pack :: proc(a: ^Int, buf: []$T, nails := 0, order := Order.LSB_Fir internal_int_unpack :: proc(a: ^Int, buf: []$T, nails := 0, order := Order.LSB_First, allocator := context.allocator) -> (err: Error) - where intrinsics.type_is_integer(T) && intrinsics.type_is_unsigned(T) && size_of(T) <= 16 { + where intrinsics.type_is_integer(T), intrinsics.type_is_unsigned(T), size_of(T) <= 16 { assert(nails >= 0 && nails < (size_of(T) * 8)) context.allocator = allocator diff --git a/core/os/os_freebsd.odin b/core/os/os_freebsd.odin index be86854dd..cdd44d301 100644 --- a/core/os/os_freebsd.odin +++ b/core/os/os_freebsd.odin @@ -159,7 +159,7 @@ blkcnt_t :: i64 blksize_t :: i32 fflags_t :: u32 -when ODIN_ARCH == .amd64 /* LP64 */ { +when ODIN_ARCH == .amd64 || ODIN_ARCH == .arm64 /* LP64 */ { time_t :: i64 } else { time_t :: i32 diff --git a/core/simd/x86/sse3.odin b/core/simd/x86/sse3.odin index cf5f3b2fa..ca19c3954 100644 --- a/core/simd/x86/sse3.odin +++ b/core/simd/x86/sse3.odin @@ -36,7 +36,7 @@ _mm_lddqu_si128 :: #force_inline proc "c" (mem_addr: ^__m128i) -> __m128i { _mm_movedup_pd :: #force_inline proc "c" (a: __m128d) -> __m128d { return simd.shuffle(a, a, 0, 0) } -@(require_results, enable_target_feature="sse3") +@(require_results, enable_target_feature="sse2,sse3") _mm_loaddup_pd :: #force_inline proc "c" (mem_addr: [^]f64) -> __m128d { return _mm_load1_pd(mem_addr) } @@ -65,4 +65,4 @@ foreign _ { hsubps :: proc(a, b: __m128) -> __m128 --- @(link_name = "llvm.x86.sse3.ldu.dq") lddqu :: proc(mem_addr: rawptr) -> i8x16 --- -}
\ No newline at end of file +} diff --git a/core/simd/x86/sse41.odin b/core/simd/x86/sse41.odin index 8c306ba4c..0b9c5986f 100644 --- a/core/simd/x86/sse41.odin +++ b/core/simd/x86/sse41.odin @@ -268,7 +268,7 @@ _mm_testnzc_si128 :: #force_inline proc "c" (a: __m128i, mask: __m128i) -> i32 { _mm_test_all_zeros :: #force_inline proc "c" (a: __m128i, mask: __m128i) -> i32 { return _mm_testz_si128(a, mask) } -@(require_results, enable_target_feature="sse4.1") +@(require_results, enable_target_feature="sse2,sse4.1") _mm_test_all_ones :: #force_inline proc "c" (a: __m128i) -> i32 { return _mm_testc_si128(a, _mm_cmpeq_epi32(a, a)) } @@ -349,4 +349,4 @@ foreign _ { ptestc :: proc(a, mask: i64x2) -> i32 --- @(link_name = "llvm.x86.sse41.ptestnzc") ptestnzc :: proc(a, mask: i64x2) -> i32 --- -}
\ No newline at end of file +} diff --git a/core/slice/slice.odin b/core/slice/slice.odin index dd8d9868a..03791e7dd 100644 --- a/core/slice/slice.odin +++ b/core/slice/slice.odin @@ -701,3 +701,39 @@ enumerated_array :: proc(ptr: ^$T) -> []intrinsics.type_elem_type(T) where intrinsics.type_is_enumerated_array(T) { return ([^]intrinsics.type_elem_type(T))(ptr)[:len(T)] } + +// Turn a `[]E` into `bit_set[E]` +// e.g.: +// bs := slice.enum_slice_to_bitset(my_flag_slice, rl.ConfigFlags) +@(require_results) +enum_slice_to_bitset :: proc(enums: []$E, $T: typeid/bit_set[E]) -> (bits: T) where intrinsics.type_is_enum(E), intrinsics.type_bit_set_elem_type(T) == E { + for v in enums { + bits |= {v} + } + return +} + +// Turn a `bit_set[E]` into a `[]E` +// e.g.: +// sl := slice.bitset_to_enum_slice(flag_buf[:], bs) +@(require_results) +bitset_to_enum_slice_with_buffer :: proc(buf: []$E, bs: $T) -> (slice: []E) where intrinsics.type_is_enum(E), intrinsics.type_bit_set_elem_type(T) == E { + count := 0 + for v in bs { + buf[count] = v + count += 1 + } + return buf[:count] +} + +// Turn a `bit_set[E]` into a `[]E`, allocates +// e.g.: +// sl := slice.bitset_to_enum_slice(bs) +@(require_results) +bitset_to_enum_slice_with_make :: proc(bs: $T, $E: typeid, allocator := context.allocator) -> (slice: []E) where intrinsics.type_is_enum(E), intrinsics.type_bit_set_elem_type(T) == E { + ones := intrinsics.count_ones(transmute(E)bs) + buf := make([]E, int(ones), allocator) + return bitset_to_enum_slice(buf, bs) +} + +bitset_to_enum_slice :: proc{bitset_to_enum_slice_with_make, bitset_to_enum_slice_with_buffer}
\ No newline at end of file diff --git a/core/sync/extended.odin b/core/sync/extended.odin index 76b7686fe..781ed816e 100644 --- a/core/sync/extended.odin +++ b/core/sync/extended.odin @@ -433,7 +433,7 @@ One_Shot_Event :: struct #no_copy { // Blocks the current thread until the event is made available with `one_shot_event_signal`. one_shot_event_wait :: proc "contextless" (e: ^One_Shot_Event) { for atomic_load_explicit(&e.state, .Acquire) == 0 { - futex_wait(&e.state, 1) + futex_wait(&e.state, 0) } } diff --git a/core/sync/futex_wasm.odin b/core/sync/futex_wasm.odin index de1013364..de88e8198 100644 --- a/core/sync/futex_wasm.odin +++ b/core/sync/futex_wasm.odin @@ -5,31 +5,49 @@ package sync import "base:intrinsics" import "core:time" +// NOTE: because `core:sync` is in the dependency chain of a lot of the core packages (mostly through `core:mem`) +// without actually calling into it much, I opted for a runtime panic instead of a compile error here. + _futex_wait :: proc "contextless" (f: ^Futex, expected: u32) -> bool { - s := intrinsics.wasm_memory_atomic_wait32((^u32)(f), expected, -1) - return s != 0 + when !intrinsics.has_target_feature("atomics") { + _panic("usage of `core:sync` requires the `-target-feature:\"atomics\"` or a `-microarch` that supports it") + } else { + s := intrinsics.wasm_memory_atomic_wait32((^u32)(f), expected, -1) + return s != 0 + } } _futex_wait_with_timeout :: proc "contextless" (f: ^Futex, expected: u32, duration: time.Duration) -> bool { - s := intrinsics.wasm_memory_atomic_wait32((^u32)(f), expected, i64(duration)) - return s != 0 - + when !intrinsics.has_target_feature("atomics") { + _panic("usage of `core:sync` requires the `-target-feature:\"atomics\"` or a `-microarch` that supports it") + } else { + s := intrinsics.wasm_memory_atomic_wait32((^u32)(f), expected, i64(duration)) + return s != 0 + } } _futex_signal :: proc "contextless" (f: ^Futex) { - loop: for { - s := intrinsics.wasm_memory_atomic_notify32((^u32)(f), 1) - if s >= 1 { - return + when !intrinsics.has_target_feature("atomics") { + _panic("usage of `core:sync` requires the `-target-feature:\"atomics\"` or a `-microarch` that supports it") + } else { + loop: for { + s := intrinsics.wasm_memory_atomic_notify32((^u32)(f), 1) + if s >= 1 { + return + } } } } _futex_broadcast :: proc "contextless" (f: ^Futex) { - loop: for { - s := intrinsics.wasm_memory_atomic_notify32((^u32)(f), ~u32(0)) - if s >= 0 { - return + when !intrinsics.has_target_feature("atomics") { + _panic("usage of `core:sync` requires the `-target-feature:\"atomics\"` or a `-microarch` that supports it") + } else { + loop: for { + s := intrinsics.wasm_memory_atomic_notify32((^u32)(f), ~u32(0)) + if s >= 0 { + return + } } } } diff --git a/core/sys/darwin/sync.odin b/core/sys/darwin/sync.odin index c76b30d6b..361b4b8b4 100644 --- a/core/sys/darwin/sync.odin +++ b/core/sys/darwin/sync.odin @@ -5,9 +5,9 @@ foreign import system "system:System.framework" // #define OS_WAIT_ON_ADDR_AVAILABILITY \ // __API_AVAILABLE(macos(14.4), ios(17.4), tvos(17.4), watchos(10.4)) when ODIN_OS == .Darwin { - when ODIN_PLATFORM_SUBTARGET == .iOS && ODIN_MINIMUM_OS_VERSION > 17_04_00 { + when ODIN_PLATFORM_SUBTARGET == .iOS && ODIN_MINIMUM_OS_VERSION >= 17_04_00 { WAIT_ON_ADDRESS_AVAILABLE :: true - } else when ODIN_MINIMUM_OS_VERSION > 14_04_00 { + } else when ODIN_MINIMUM_OS_VERSION >= 14_04_00 { WAIT_ON_ADDRESS_AVAILABLE :: true } else { WAIT_ON_ADDRESS_AVAILABLE :: false diff --git a/core/sys/linux/bits.odin b/core/sys/linux/bits.odin index 4db689643..cfae06013 100644 --- a/core/sys/linux/bits.odin +++ b/core/sys/linux/bits.odin @@ -153,22 +153,41 @@ Errno :: enum i32 { Open_Flags_Bits :: enum { WRONLY = 0, RDWR = 1, - CREAT = 8, - EXCL = 9, - NOCTTY = 10, - TRUNC = 11, - APPEND = 12, - NONBLOCK = 14, - DSYNC = 16, - ASYNC = 17, - DIRECT = 18, - LARGEFILE = 20, - DIRECTORY = 21, - NOFOLLOW = 22, - NOATIME = 24, - CLOEXEC = 25, - PATH = 28, -} + CREAT = 6, + EXCL = 7, + NOCTTY = 8, + TRUNC = 9, + APPEND = 10, + NONBLOCK = 11, + DSYNC = 12, + ASYNC = 13, + DIRECT = 14, + LARGEFILE = 15, + DIRECTORY = 16, + NOFOLLOW = 17, + NOATIME = 18, + CLOEXEC = 19, + PATH = 21, +} + +// https://github.com/torvalds/linux/blob/7367539ad4b0f8f9b396baf02110962333719a48/include/uapi/asm-generic/fcntl.h#L19 +#assert(1 << uint(Open_Flags_Bits.WRONLY) == 0o0000000_1) +#assert(1 << uint(Open_Flags_Bits.RDWR) == 0o0000000_2) +#assert(1 << uint(Open_Flags_Bits.CREAT) == 0o00000_100) +#assert(1 << uint(Open_Flags_Bits.EXCL) == 0o00000_200) +#assert(1 << uint(Open_Flags_Bits.NOCTTY) == 0o00000_400) +#assert(1 << uint(Open_Flags_Bits.TRUNC) == 0o0000_1000) +#assert(1 << uint(Open_Flags_Bits.APPEND) == 0o0000_2000) +#assert(1 << uint(Open_Flags_Bits.NONBLOCK) == 0o0000_4000) +#assert(1 << uint(Open_Flags_Bits.DSYNC) == 0o000_10000) +#assert(1 << uint(Open_Flags_Bits.ASYNC) == 0o000_20000) +#assert(1 << uint(Open_Flags_Bits.DIRECT) == 0o000_40000) +#assert(1 << uint(Open_Flags_Bits.LARGEFILE) == 0o00_100000) +#assert(1 << uint(Open_Flags_Bits.DIRECTORY) == 0o00_200000) +#assert(1 << uint(Open_Flags_Bits.NOFOLLOW) == 0o00_400000) +#assert(1 << uint(Open_Flags_Bits.NOATIME) == 0o0_1000000) +#assert(1 << uint(Open_Flags_Bits.CLOEXEC) == 0o0_2000000) +#assert(1 << uint(Open_Flags_Bits.PATH) == 0o_10000000) /* Bits for FD_Flags bitset diff --git a/core/sys/linux/helpers.odin b/core/sys/linux/helpers.odin index 69c648bf1..75fdd586e 100644 --- a/core/sys/linux/helpers.odin +++ b/core/sys/linux/helpers.odin @@ -26,7 +26,7 @@ where @(private) syscall2 :: #force_inline proc "contextless" (nr: uintptr,p1: $T1, p2: $T2) -> int where - size_of(p1) <= size_of(uintptr) && + size_of(p1) <= size_of(uintptr), size_of(p2) <= size_of(uintptr) { return cast(int) intrinsics.syscall(nr, @@ -36,8 +36,8 @@ where @(private) syscall3 :: #force_inline proc "contextless" (nr: uintptr, p1: $T1, p2: $T2, p3: $T3) -> int where - size_of(p1) <= size_of(uintptr) && - size_of(p2) <= size_of(uintptr) && + size_of(p1) <= size_of(uintptr), + size_of(p2) <= size_of(uintptr), size_of(p3) <= size_of(uintptr) { return cast(int) intrinsics.syscall(nr, @@ -49,9 +49,9 @@ where @(private) syscall4 :: #force_inline proc "contextless" (nr: uintptr, p1: $T1, p2: $T2, p3: $T3, p4: $T4) -> int where - size_of(p1) <= size_of(uintptr) && - size_of(p2) <= size_of(uintptr) && - size_of(p3) <= size_of(uintptr) && + size_of(p1) <= size_of(uintptr), + size_of(p2) <= size_of(uintptr), + size_of(p3) <= size_of(uintptr), size_of(p4) <= size_of(uintptr) { return cast(int) intrinsics.syscall(nr, @@ -64,10 +64,10 @@ where @(private) syscall5 :: #force_inline proc "contextless" (nr: uintptr, p1: $T1, p2: $T2, p3: $T3, p4: $T4, p5: $T5) -> int where - size_of(p1) <= size_of(uintptr) && - size_of(p2) <= size_of(uintptr) && - size_of(p3) <= size_of(uintptr) && - size_of(p4) <= size_of(uintptr) && + size_of(p1) <= size_of(uintptr), + size_of(p2) <= size_of(uintptr), + size_of(p3) <= size_of(uintptr), + size_of(p4) <= size_of(uintptr), size_of(p5) <= size_of(uintptr) { return cast(int) intrinsics.syscall(nr, @@ -81,11 +81,11 @@ where @(private) syscall6 :: #force_inline proc "contextless" (nr: uintptr, p1: $T1, p2: $T2, p3: $T3, p4: $T4, p5: $T5, p6: $T6) -> int where - size_of(p1) <= size_of(uintptr) && - size_of(p2) <= size_of(uintptr) && - size_of(p3) <= size_of(uintptr) && - size_of(p4) <= size_of(uintptr) && - size_of(p5) <= size_of(uintptr) && + size_of(p1) <= size_of(uintptr), + size_of(p2) <= size_of(uintptr), + size_of(p3) <= size_of(uintptr), + size_of(p4) <= size_of(uintptr), + size_of(p5) <= size_of(uintptr), size_of(p6) <= size_of(uintptr) { return cast(int) intrinsics.syscall(nr, diff --git a/core/sys/linux/sys.odin b/core/sys/linux/sys.odin index 63fb3b776..413c8742b 100644 --- a/core/sys/linux/sys.odin +++ b/core/sys/linux/sys.odin @@ -2314,7 +2314,7 @@ futex :: proc { */ epoll_create :: proc(size: i32 = 1) -> (Fd, Errno) { when ODIN_ARCH != .arm64 { - ret := syscall(SYS_epoll_create) + ret := syscall(SYS_epoll_create, i32(1)) return errno_unwrap(ret, Fd) } else { ret := syscall(SYS_epoll_create1, i32(0)) diff --git a/core/thread/thread_unix.odin b/core/thread/thread_unix.odin index e25d3541c..36ee035a5 100644 --- a/core/thread/thread_unix.odin +++ b/core/thread/thread_unix.odin @@ -25,7 +25,7 @@ _create :: proc(procedure: Thread_Proc, priority: Thread_Priority) -> ^Thread { when ODIN_OS != .Darwin { // We need to give the thread a moment to start up before we enable cancellation. - can_set_thread_cancel_state := unix.pthread_setcancelstate(unix.PTHREAD_CANCEL_DISABLE, nil) == 0 + can_set_thread_cancel_state := unix.pthread_setcancelstate(unix.PTHREAD_CANCEL_ENABLE, nil) == 0 } sync.lock(&t.mutex) @@ -40,7 +40,7 @@ _create :: proc(procedure: Thread_Proc, priority: Thread_Priority) -> ^Thread { // Enable thread's cancelability. if can_set_thread_cancel_state { unix.pthread_setcanceltype (unix.PTHREAD_CANCEL_ASYNCHRONOUS, nil) - unix.pthread_setcancelstate(unix.PTHREAD_CANCEL_DISABLE, nil) + unix.pthread_setcancelstate(unix.PTHREAD_CANCEL_ENABLE, nil) } } diff --git a/examples/demo/demo.odin b/examples/demo/demo.odin index a62c11310..39345e39c 100644 --- a/examples/demo/demo.odin +++ b/examples/demo/demo.odin @@ -48,7 +48,7 @@ the_basics :: proc() { // os.args holds the path to the current executable and any arguments passed to it. if len(os.args) == 1 { fmt.printf("Hellope from %v.\n", os.args[0]) - } else { + } else if len(os.args) > 2 { fmt.printf("%v, %v! from %v.\n", os.args[1], os.args[2], os.args[0]) } diff --git a/misc/featuregen/README.md b/misc/featuregen/README.md new file mode 100644 index 000000000..22a798cca --- /dev/null +++ b/misc/featuregen/README.md @@ -0,0 +1,28 @@ +# Featuregen + +This directory contains a python and CPP script that generates the needed information +for features regarding microarchitecture and target features of the compiler. + +It is not pretty! But LLVM has no way to query this information with their C API. + +It generates these globals (intended for `src/build_settings.cpp`: + +- `target_microarch_list`: an array of strings indexed by the architecture, each string is a comma-seperated list of microarchitectures available on that architecture +- `target_features_list`: an array of strings indexed by the architecture, each string is a comma-seperated list of target features available on that architecture +- `target_microarch_counts`: an array of ints indexed by the architecture, each int represents the amount of microarchitectures available on that target, intended for easier iteration of the next global +- `microarch_features_list`: an array of a tuple like struct where the first string is a microarchitecture and the second is a comma-seperated list of all features that are enabled by default for it + +In order to get the default features for a microarchitecture there is a small CPP program that takes +a target triple and microarchitecture and spits out the default features, this is then parsed by the python script. + +This should be ran each time we update LLVM to stay in sync. + +If there are minor differences (like the Odin user using LLVM 14 and this table being generated on LLVM 17) it +does not impact much at all, the only thing it will do is make LLVM print a message that the feature is ignored (if it was added between 14 and 17 in this case). + +## Usage + +1. Make sure the table of architectures at the top of the python script is up-to-date (the triple can be any valid triple for the architecture) +1. `./build.sh` +1. `python3 featuregen.py` +1. Copy the output into `src/build_settings.cpp` diff --git a/misc/featuregen/featuregen.cpp b/misc/featuregen/featuregen.cpp new file mode 100644 index 000000000..a1d00ab31 --- /dev/null +++ b/misc/featuregen/featuregen.cpp @@ -0,0 +1,37 @@ +#include <llvm/MC/MCSubtargetInfo.h> +#include <llvm/MC/TargetRegistry.h> +#include <llvm/Support/raw_ostream.h> +#include <llvm/ADT/ArrayRef.h> +#include <llvm/Support/InitLLVM.h> +#include <llvm/Support/TargetSelect.h> + +// Dumps the default set of supported features for the given microarch. +int main(int argc, char **argv) { + if (argc < 3) { + llvm::errs() << "Error: first arg should be triple, second should be microarch\n"; + return 1; + } + + llvm::InitializeAllTargets(); + llvm::InitializeAllTargetMCs(); + + std::string error; + const llvm::Target* target = llvm::TargetRegistry::lookupTarget(argv[1], error); + + if (!target) { + llvm::errs() << "Error: " << error << "\n"; + return 1; + } + + auto STI = target->createMCSubtargetInfo(argv[1], argv[2], ""); + + std::string plus = "+"; + llvm::ArrayRef<llvm::SubtargetFeatureKV> features = STI->getAllProcessorFeatures(); + for (const auto& feature : features) { + if (STI->checkFeatures(plus + feature.Key)) { + llvm::outs() << feature.Key << "\n"; + } + } + + return 0; +} diff --git a/misc/featuregen/featuregen.py b/misc/featuregen/featuregen.py new file mode 100644 index 000000000..da4cc68f5 --- /dev/null +++ b/misc/featuregen/featuregen.py @@ -0,0 +1,116 @@ +import subprocess +import tempfile +import os +import sys + +archs = [ + ("amd64", "linux_amd64", "x86_64-pc-linux-gnu", [], []), + ("i386", "linux_i386", "i386-pc-linux-gnu", [], []), + ("arm32", "linux_arm32", "arm-linux-gnu", [], []), + ("arm64", "linux_arm64", "aarch64-linux-elf", [], []), + ("wasm32", "js_wasm32", "wasm32-js-js", [], []), + ("wasm64p32", "js_wasm64p32","wasm32-js-js", [], []), +]; + +SEEKING_CPUS = 0 +PARSING_CPUS = 1 +PARSING_FEATURES = 2 + +with tempfile.NamedTemporaryFile(suffix=".odin", delete=True) as temp_file: + temp_file.write(b"package main\n") + + for arch, target, triple, cpus, features in archs: + cmd = ["odin", "build", temp_file.name, "-file", "-build-mode:llvm", "-out:temp", "-target-features:\"help\"", f"-target:\"{target}\""] + process = subprocess.Popen(cmd, stderr=subprocess.PIPE, text=True) + + state = SEEKING_CPUS + for line in process.stderr: + + if state == SEEKING_CPUS: + if line == "Available CPUs for this target:\n": + state = PARSING_CPUS + + elif state == PARSING_CPUS: + if line == "Available features for this target:\n": + state = PARSING_FEATURES + continue + + parts = line.split(" -", maxsplit=1) + if len(parts) < 2: + continue + + cpu = parts[0].strip() + cpus.append(cpu) + + elif state == PARSING_FEATURES: + if line == "\n" and len(features) > 0: + break + + parts = line.split(" -", maxsplit=1) + if len(parts) < 2: + continue + + feature = parts[0].strip() + features.append(feature) + + process.wait() + if process.returncode != 0: + print(f"odin build returned with non-zero exit code {process.returncode}") + sys.exit(1) + + os.remove("temp.ll") + +def print_default_features(triple, microarch): + cmd = ["./featuregen", triple, microarch] + process = subprocess.Popen(cmd, stdout=subprocess.PIPE, text=True) + first = True + for line in process.stdout: + print("" if first else ",", line.strip(), sep="", end="") + first = False + process.wait() + if process.returncode != 0: + print(f"featuregen returned with non-zero exit code {process.returncode}") + sys.exit(1) + +print("// Generated with the featuregen script in `misc/featuregen`") +print("gb_global String target_microarch_list[TargetArch_COUNT] = {") +print("\t// TargetArch_Invalid:") +print('\tstr_lit(""),') +for arch, target, triple, cpus, features in archs: + print(f"\t// TargetArch_{arch}:") + print(f'\tstr_lit("{','.join(cpus)}"),') +print("};") + +print("") + +print("// Generated with the featuregen script in `misc/featuregen`") +print("gb_global String target_features_list[TargetArch_COUNT] = {") +print("\t// TargetArch_Invalid:") +print('\tstr_lit(""),') +for arch, target, triple, cpus, features in archs: + print(f"\t// TargetArch_{arch}:") + print(f'\tstr_lit("{','.join(features)}"),') +print("};") + +print("") + +print("// Generated with the featuregen script in `misc/featuregen`") +print("gb_global int target_microarch_counts[TargetArch_COUNT] = {") +print("\t// TargetArch_Invalid:") +print("\t0,") +for arch, target, triple, cpus, feature in archs: + print(f"\t// TargetArch_{arch}:") + print(f"\t{len(cpus)},") +print("};") + +print("") + +print("// Generated with the featuregen script in `misc/featuregen`") +print("gb_global MicroarchFeatureList microarch_features_list[] = {") +for arch, target, triple, cpus, features in archs: + print(f"\t// TargetArch_{arch}:") + for cpu in cpus: + print(f'\t{{ str_lit("{cpu}"), str_lit("', end="") + print_default_features(triple, cpu) + print('") },') +print("};") diff --git a/src/build_settings.cpp b/src/build_settings.cpp index 77f16402b..dca57f73a 100644 --- a/src/build_settings.cpp +++ b/src/build_settings.cpp @@ -72,6 +72,11 @@ enum Windows_Subsystem : u8 { Windows_Subsystem_COUNT, }; +struct MicroarchFeatureList { + String microarch; + String features; +}; + gb_global String target_os_names[TargetOs_COUNT] = { str_lit(""), str_lit("windows"), @@ -99,21 +104,467 @@ gb_global String target_arch_names[TargetArch_COUNT] = { str_lit("wasm64p32"), }; +// Generated with the featuregen script in `misc/featuregen` gb_global String target_microarch_list[TargetArch_COUNT] = { - // TargetArch_Invalid, - str_lit("Invalid!"), - // TargetArch_amd64, - str_lit("alderlake,amdfam10,athlon-fx,athlon64,athlon64-sse3,atom_sse4_2,atom_sse4_2_movbe,barcelona,bdver1,bdver2,bdver3,bdver4,broadwell,btver1,btver2,cannonlake,cascadelake,cooperlake,core-avx-i,core-avx2,core2,core_2_duo_sse4_1,core_2_duo_ssse3,core_2nd_gen_avx,core_3rd_gen_avx,core_4th_gen_avx,core_4th_gen_avx_tsx,core_5th_gen_avx,core_5th_gen_avx_tsx,core_aes_pclmulqdq,core_i7_sse4_2,corei7,corei7-avx,generic,goldmont,goldmont-plus,goldmont_plus,grandridge,graniterapids,graniterapids-d,graniterapids_d,haswell,icelake-client,icelake-server,icelake_client,icelake_server,ivybridge,k8,k8-sse3,knl,knm,meteorlake,mic_avx512,native,nehalem,nocona,opteron,opteron-sse3,penryn,raptorlake,rocketlake,sandybridge,sapphirerapids,sierraforest,silvermont,skx,skylake,skylake-avx512,skylake_avx512,slm,tigerlake,tremont,westmere,x86-64,x86-64-v2,x86-64-v3,x86-64-v4,znver1,znver2,znver3,znver4"), - // TargetArch_i386, - str_lit("athlon,athlon-4,athlon-mp,athlon-tbird,athlon-xp,atom,bonnell,c3,c3-2,generic,geode,i386,i486,i586,i686,k6,k6-2,k6-3,lakemont,native,pentium,pentium-m,pentium-mmx,pentium2,pentium3,pentium3m,pentium4,pentium4m,pentium_4,pentium_4_sse3,pentium_ii,pentium_iii,pentium_iii_no_xmm_regs,pentium_m,pentium_mmx,pentium_pro,pentiumpro,prescott,winchip-c6,winchip2,yonah"), - // TargetArch_arm32, - str_lit("arm1020e,arm1020t,arm1022e,arm10e,arm10tdmi,arm1136j-s,arm1136jf-s,arm1156t2-s,arm1156t2f-s,arm1176jz-s,arm1176jzf-s,arm710t,arm720t,arm7tdmi,arm7tdmi-s,arm8,arm810,arm9,arm920,arm920t,arm922t,arm926ej-s,arm940t,arm946e-s,arm966e-s,arm968e-s,arm9e,arm9tdmi,cortex-a12,cortex-a15,cortex-a17,cortex-a32,cortex-a35,cortex-a5,cortex-a53,cortex-a55,cortex-a57,cortex-a7,cortex-a710,cortex-a72,cortex-a73,cortex-a75,cortex-a76,cortex-a76ae,cortex-a77,cortex-a78,cortex-a78c,cortex-a8,cortex-a9,cortex-m0,cortex-m0plus,cortex-m1,cortex-m23,cortex-m3,cortex-m33,cortex-m35p,cortex-m4,cortex-m55,cortex-m7,cortex-m85,cortex-r4,cortex-r4f,cortex-r5,cortex-r52,cortex-r7,cortex-r8,cortex-x1,cortex-x1c,cyclone,ep9312,exynos-m3,exynos-m4,exynos-m5,generic,iwmmxt,krait,kryo,mpcore,mpcorenovfp,native,neoverse-n1,neoverse-n2,neoverse-v1,sc000,sc300,strongarm,strongarm110,strongarm1100,strongarm1110,swift,xscale"), - // TargetArch_arm64, - str_lit("a64fx,ampere1,ampere1a,apple-a10,apple-a11,apple-a12,apple-a13,apple-a14,apple-a15,apple-a16,apple-a7,apple-a8,apple-a9,apple-latest,apple-m1,apple-m2,apple-s4,apple-s5,carmel,cortex-a34,cortex-a35,cortex-a510,cortex-a53,cortex-a55,cortex-a57,cortex-a65,cortex-a65ae,cortex-a710,cortex-a715,cortex-a72,cortex-a73,cortex-a75,cortex-a76,cortex-a76ae,cortex-a77,cortex-a78,cortex-a78c,cortex-r82,cortex-x1,cortex-x1c,cortex-x2,cortex-x3,cyclone,exynos-m3,exynos-m4,exynos-m5,falkor,generic,kryo,native,neoverse-512tvb,neoverse-e1,neoverse-n1,neoverse-n2,neoverse-v1,neoverse-v2,saphira,thunderx,thunderx2t99,thunderx3t110,thunderxt81,thunderxt83,thunderxt88,tsv110"), - // TargetArch_wasm32, - str_lit("generic"), - // TargetArch_wasm64p32, - str_lit("generic"), + // TargetArch_Invalid: + str_lit(""), + // TargetArch_amd64: + str_lit("alderlake,amdfam10,athlon,athlon-4,athlon-fx,athlon-mp,athlon-tbird,athlon-xp,athlon64,athlon64-sse3,atom,atom_sse4_2,atom_sse4_2_movbe,barcelona,bdver1,bdver2,bdver3,bdver4,bonnell,broadwell,btver1,btver2,c3,c3-2,cannonlake,cascadelake,cooperlake,core-avx-i,core-avx2,core2,core_2_duo_sse4_1,core_2_duo_ssse3,core_2nd_gen_avx,core_3rd_gen_avx,core_4th_gen_avx,core_4th_gen_avx_tsx,core_5th_gen_avx,core_5th_gen_avx_tsx,core_aes_pclmulqdq,core_i7_sse4_2,corei7,corei7-avx,emeraldrapids,generic,geode,goldmont,goldmont-plus,goldmont_plus,grandridge,graniterapids,graniterapids-d,graniterapids_d,haswell,i386,i486,i586,i686,icelake-client,icelake-server,icelake_client,icelake_server,ivybridge,k6,k6-2,k6-3,k8,k8-sse3,knl,knm,lakemont,meteorlake,mic_avx512,nehalem,nocona,opteron,opteron-sse3,penryn,pentium,pentium-m,pentium-mmx,pentium2,pentium3,pentium3m,pentium4,pentium4m,pentium_4,pentium_4_sse3,pentium_ii,pentium_iii,pentium_iii_no_xmm_regs,pentium_m,pentium_mmx,pentium_pro,pentiumpro,prescott,raptorlake,rocketlake,sandybridge,sapphirerapids,sierraforest,silvermont,skx,skylake,skylake-avx512,skylake_avx512,slm,tigerlake,tremont,westmere,winchip-c6,winchip2,x86-64,x86-64-v2,x86-64-v3,x86-64-v4,yonah,znver1,znver2,znver3,znver4"), + // TargetArch_i386: + str_lit("alderlake,amdfam10,athlon,athlon-4,athlon-fx,athlon-mp,athlon-tbird,athlon-xp,athlon64,athlon64-sse3,atom,atom_sse4_2,atom_sse4_2_movbe,barcelona,bdver1,bdver2,bdver3,bdver4,bonnell,broadwell,btver1,btver2,c3,c3-2,cannonlake,cascadelake,cooperlake,core-avx-i,core-avx2,core2,core_2_duo_sse4_1,core_2_duo_ssse3,core_2nd_gen_avx,core_3rd_gen_avx,core_4th_gen_avx,core_4th_gen_avx_tsx,core_5th_gen_avx,core_5th_gen_avx_tsx,core_aes_pclmulqdq,core_i7_sse4_2,corei7,corei7-avx,emeraldrapids,generic,geode,goldmont,goldmont-plus,goldmont_plus,grandridge,graniterapids,graniterapids-d,graniterapids_d,haswell,i386,i486,i586,i686,icelake-client,icelake-server,icelake_client,icelake_server,ivybridge,k6,k6-2,k6-3,k8,k8-sse3,knl,knm,lakemont,meteorlake,mic_avx512,nehalem,nocona,opteron,opteron-sse3,penryn,pentium,pentium-m,pentium-mmx,pentium2,pentium3,pentium3m,pentium4,pentium4m,pentium_4,pentium_4_sse3,pentium_ii,pentium_iii,pentium_iii_no_xmm_regs,pentium_m,pentium_mmx,pentium_pro,pentiumpro,prescott,raptorlake,rocketlake,sandybridge,sapphirerapids,sierraforest,silvermont,skx,skylake,skylake-avx512,skylake_avx512,slm,tigerlake,tremont,westmere,winchip-c6,winchip2,x86-64,x86-64-v2,x86-64-v3,x86-64-v4,yonah,znver1,znver2,znver3,znver4"), + // TargetArch_arm32: + str_lit("arm1020e,arm1020t,arm1022e,arm10e,arm10tdmi,arm1136j-s,arm1136jf-s,arm1156t2-s,arm1156t2f-s,arm1176jz-s,arm1176jzf-s,arm710t,arm720t,arm7tdmi,arm7tdmi-s,arm8,arm810,arm9,arm920,arm920t,arm922t,arm926ej-s,arm940t,arm946e-s,arm966e-s,arm968e-s,arm9e,arm9tdmi,cortex-a12,cortex-a15,cortex-a17,cortex-a32,cortex-a35,cortex-a5,cortex-a53,cortex-a55,cortex-a57,cortex-a7,cortex-a710,cortex-a72,cortex-a73,cortex-a75,cortex-a76,cortex-a76ae,cortex-a77,cortex-a78,cortex-a78c,cortex-a8,cortex-a9,cortex-m0,cortex-m0plus,cortex-m1,cortex-m23,cortex-m3,cortex-m33,cortex-m35p,cortex-m4,cortex-m55,cortex-m7,cortex-m85,cortex-r4,cortex-r4f,cortex-r5,cortex-r52,cortex-r7,cortex-r8,cortex-x1,cortex-x1c,cyclone,ep9312,exynos-m3,exynos-m4,exynos-m5,generic,iwmmxt,krait,kryo,mpcore,mpcorenovfp,neoverse-n1,neoverse-n2,neoverse-v1,sc000,sc300,strongarm,strongarm110,strongarm1100,strongarm1110,swift,xscale"), + // TargetArch_arm64: + str_lit("a64fx,ampere1,ampere1a,apple-a10,apple-a11,apple-a12,apple-a13,apple-a14,apple-a15,apple-a16,apple-a7,apple-a8,apple-a9,apple-latest,apple-m1,apple-m2,apple-s4,apple-s5,carmel,cortex-a34,cortex-a35,cortex-a510,cortex-a53,cortex-a55,cortex-a57,cortex-a65,cortex-a65ae,cortex-a710,cortex-a715,cortex-a72,cortex-a73,cortex-a75,cortex-a76,cortex-a76ae,cortex-a77,cortex-a78,cortex-a78c,cortex-r82,cortex-x1,cortex-x1c,cortex-x2,cortex-x3,cyclone,exynos-m3,exynos-m4,exynos-m5,falkor,generic,kryo,neoverse-512tvb,neoverse-e1,neoverse-n1,neoverse-n2,neoverse-v1,neoverse-v2,saphira,thunderx,thunderx2t99,thunderx3t110,thunderxt81,thunderxt83,thunderxt88,tsv110"), + // TargetArch_wasm32: + str_lit("bleeding-edge,generic,mvp"), + // TargetArch_wasm64p32: + str_lit("bleeding-edge,generic,mvp"), +}; + +// Generated with the featuregen script in `misc/featuregen` +gb_global String target_features_list[TargetArch_COUNT] = { + // TargetArch_Invalid: + str_lit(""), + // TargetArch_amd64: + str_lit("16bit-mode,32bit-mode,3dnow,3dnowa,64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512er,avx512f,avx512fp16,avx512ifma,avx512pf,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vp2intersect,avx512vpopcntdq,avxifma,avxneconvert,avxvnni,avxvnniint16,avxvnniint8,bmi,bmi2,branchfusion,cldemote,clflushopt,clwb,clzero,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,ermsb,f16c,false-deps-getmant,false-deps-lzcnt-tzcnt,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-popcnt,false-deps-range,fast-11bytenop,fast-15bytenop,fast-7bytenop,fast-bextr,fast-gather,fast-hops,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fast-vector-shift-masks,faster-shift-than-shuffle,fma,fma4,fsgsbase,fsrm,fxsr,gfni,harden-sls-ijmp,harden-sls-ret,hreset,idivl-to-divb,idivq-to-divl,invpcid,kl,lea-sp,lea-uses-ag,lvi-cfi,lvi-load-hardening,lwp,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,mwaitx,no-bypass-delay,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pad-short-functions,pclmul,pconfig,pku,popcnt,prefer-128-bit,prefer-256-bit,prefer-mask-registers,prefer-movmsk-over-vtest,prefer-no-gather,prefer-no-scatter,prefetchi,prefetchwt1,prfchw,ptwrite,raoint,rdpid,rdpru,rdrnd,rdseed,retpoline,retpoline-external-thunk,retpoline-indirect-branches,retpoline-indirect-calls,rtm,sahf,sbb-dep-breaking,serialize,seses,sgx,sha,sha512,shstk,slow-3ops-lea,slow-incdec,slow-lea,slow-pmaddwd,slow-pmulld,slow-shld,slow-two-mem-ops,slow-unaligned-mem-16,slow-unaligned-mem-32,sm3,sm4,soft-float,sse,sse-unaligned-mem,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tagged-globals,tbm,tsxldtrk,tuning-fast-imm-vector-shift,uintr,use-glm-div-sqrt-costs,use-slm-arith-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,widekl,x87,xop,xsave,xsavec,xsaveopt,xsaves"), + // TargetArch_i386: + str_lit("16bit-mode,32bit-mode,3dnow,3dnowa,64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512er,avx512f,avx512fp16,avx512ifma,avx512pf,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vp2intersect,avx512vpopcntdq,avxifma,avxneconvert,avxvnni,avxvnniint16,avxvnniint8,bmi,bmi2,branchfusion,cldemote,clflushopt,clwb,clzero,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,ermsb,f16c,false-deps-getmant,false-deps-lzcnt-tzcnt,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-popcnt,false-deps-range,fast-11bytenop,fast-15bytenop,fast-7bytenop,fast-bextr,fast-gather,fast-hops,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fast-vector-shift-masks,faster-shift-than-shuffle,fma,fma4,fsgsbase,fsrm,fxsr,gfni,harden-sls-ijmp,harden-sls-ret,hreset,idivl-to-divb,idivq-to-divl,invpcid,kl,lea-sp,lea-uses-ag,lvi-cfi,lvi-load-hardening,lwp,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,mwaitx,no-bypass-delay,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pad-short-functions,pclmul,pconfig,pku,popcnt,prefer-128-bit,prefer-256-bit,prefer-mask-registers,prefer-movmsk-over-vtest,prefer-no-gather,prefer-no-scatter,prefetchi,prefetchwt1,prfchw,ptwrite,raoint,rdpid,rdpru,rdrnd,rdseed,retpoline,retpoline-external-thunk,retpoline-indirect-branches,retpoline-indirect-calls,rtm,sahf,sbb-dep-breaking,serialize,seses,sgx,sha,sha512,shstk,slow-3ops-lea,slow-incdec,slow-lea,slow-pmaddwd,slow-pmulld,slow-shld,slow-two-mem-ops,slow-unaligned-mem-16,slow-unaligned-mem-32,sm3,sm4,soft-float,sse,sse-unaligned-mem,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tagged-globals,tbm,tsxldtrk,tuning-fast-imm-vector-shift,uintr,use-glm-div-sqrt-costs,use-slm-arith-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,widekl,x87,xop,xsave,xsavec,xsaveopt,xsaves"), + // TargetArch_arm32: + str_lit("32bit,8msecext,a12,a15,a17,a32,a35,a5,a53,a55,a57,a7,a72,a73,a75,a76,a77,a78c,a8,a9,aapcs-frame-chain,aapcs-frame-chain-leaf,aclass,acquire-release,aes,armv4,armv4t,armv5t,armv5te,armv5tej,armv6,armv6-m,armv6j,armv6k,armv6kz,armv6s-m,armv6t2,armv7-a,armv7-m,armv7-r,armv7e-m,armv7k,armv7s,armv7ve,armv8-a,armv8-m.base,armv8-m.main,armv8-r,armv8.1-a,armv8.1-m.main,armv8.2-a,armv8.3-a,armv8.4-a,armv8.5-a,armv8.6-a,armv8.7-a,armv8.8-a,armv8.9-a,armv9-a,armv9.1-a,armv9.2-a,armv9.3-a,armv9.4-a,atomics-32,avoid-movs-shop,avoid-partial-cpsr,bf16,big-endian-instructions,cde,cdecp0,cdecp1,cdecp2,cdecp3,cdecp4,cdecp5,cdecp6,cdecp7,cheap-predicable-cpsr,clrbhb,cortex-a710,cortex-a78,cortex-x1,cortex-x1c,crc,crypto,d32,db,dfb,disable-postra-scheduler,dont-widen-vmovs,dotprod,dsp,execute-only,expand-fp-mlx,exynos,fix-cmse-cve-2021-35465,fix-cortex-a57-aes-1742098,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp16fml,fp64,fpao,fpregs,fpregs16,fpregs64,fullfp16,fuse-aes,fuse-literals,harden-sls-blr,harden-sls-nocomdat,harden-sls-retbr,hwdiv,hwdiv-arm,i8mm,iwmmxt,iwmmxt2,krait,kryo,lob,long-calls,loop-align,m3,m7,mclass,mp,muxed-units,mve,mve.fp,mve1beat,mve2beat,mve4beat,nacl-trap,neon,neon-fpmovs,neonfp,neoverse-v1,no-branch-predictor,no-bti-at-return-twice,no-movt,no-neg-immediates,noarm,nonpipelined-vfp,pacbti,perfmon,prefer-ishst,prefer-vmovsr,prof-unpr,r4,r5,r52,r7,ras,rclass,read-tp-tpidrprw,read-tp-tpidruro,read-tp-tpidrurw,reserve-r9,ret-addr-stack,sb,sha2,slow-fp-brcc,slow-load-D-subreg,slow-odd-reg,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,soft-float,splat-vfp-neon,strict-align,swift,thumb-mode,thumb2,trustzone,use-mipipeliner,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.1m.main,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8.7a,v8.8a,v8.9a,v8m,v8m.main,v9.1a,v9.2a,v9.3a,v9.4a,v9a,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vldn-align,vmlx-forwarding,vmlx-hazards,wide-stride-vfp,xscale,zcz"), + // TargetArch_arm64: + str_lit("CONTEXTIDREL2,a35,a510,a53,a55,a57,a64fx,a65,a710,a715,a72,a73,a75,a76,a77,a78,a78c,aes,aggressive-fma,all,alternate-sextload-cvt-f32-pattern,altnzcv,am,ampere1,ampere1a,amvs,apple-a10,apple-a11,apple-a12,apple-a13,apple-a14,apple-a15,apple-a16,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,ascend-store-address,b16b16,balance-fp-ops,bf16,brbe,bti,call-saved-x10,call-saved-x11,call-saved-x12,call-saved-x13,call-saved-x14,call-saved-x15,call-saved-x18,call-saved-x8,call-saved-x9,carmel,ccdp,ccidx,ccpp,chk,clrbhb,cmp-bcc-fusion,complxnum,cortex-r82,cortex-x1,cortex-x2,cortex-x3,crc,crypto,cssc,custom-cheap-as-move,d128,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,enable-select-opt,ete,exynos-cheap-as-move,exynosm3,exynosm4,f32mm,f64mm,falkor,fgt,fix-cortex-a53-835769,flagm,fmv,force-32bit-jump-tables,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-addsub-2reg-const1,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,gcs,harden-sls-blr,harden-sls-nocomdat,harden-sls-retbr,hbc,hcx,i8mm,ite,jsconv,kryo,lor,ls64,lse,lse128,lse2,lsl-fast,mec,mops,mpam,mte,neon,neoverse512tvb,neoversee1,neoversen1,neoversen2,neoversev1,neoversev2,nmi,no-bti-at-return-twice,no-neg-immediates,no-sve-fp-ld1r,no-zcz-fp,nv,outline-atomics,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,prfm-slc-target,rand,ras,rasv2,rcpc,rcpc-immo,rcpc3,rdm,reserve-x1,reserve-x10,reserve-x11,reserve-x12,reserve-x13,reserve-x14,reserve-x15,reserve-x18,reserve-x2,reserve-x20,reserve-x21,reserve-x22,reserve-x23,reserve-x24,reserve-x25,reserve-x26,reserve-x27,reserve-x28,reserve-x3,reserve-x30,reserve-x4,reserve-x5,reserve-x6,reserve-x7,reserve-x9,rme,saphira,sb,sel2,sha2,sha3,slow-misaligned-128store,slow-paired-128,slow-strqro-store,sm4,sme,sme-f16f16,sme-f64f64,sme-i16i64,sme2,sme2p1,spe,spe-eef,specres2,specrestrict,ssbs,strict-align,sve,sve2,sve2-aes,sve2-bitperm,sve2-sha3,sve2-sm4,sve2p1,tagged-globals,the,thunderx,thunderx2t99,thunderx3t110,thunderxt81,thunderxt83,thunderxt88,tlb-rmi,tme,tpidr-el1,tpidr-el2,tpidr-el3,tpidrro-el0,tracev8.4,trbe,tsv110,uaops,use-experimental-zeroing-pseudos,use-postra-scheduler,use-reciprocal-square-root,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8.7a,v8.8a,v8.9a,v8a,v8r,v9.1a,v9.2a,v9.3a,v9.4a,v9a,vh,wfxt,xs,zcm,zcz,zcz-fp-workaround,zcz-gp"), + // TargetArch_wasm32: + str_lit("atomics,bulk-memory,exception-handling,extended-const,multivalue,mutable-globals,nontrapping-fptoint,reference-types,relaxed-simd,sign-ext,simd128,tail-call"), + // TargetArch_wasm64p32: + str_lit("atomics,bulk-memory,exception-handling,extended-const,multivalue,mutable-globals,nontrapping-fptoint,reference-types,relaxed-simd,sign-ext,simd128,tail-call"), +}; + +// Generated with the featuregen script in `misc/featuregen` +gb_global int target_microarch_counts[TargetArch_COUNT] = { + // TargetArch_Invalid: + 0, + // TargetArch_amd64: + 120, + // TargetArch_i386: + 120, + // TargetArch_arm32: + 90, + // TargetArch_arm64: + 63, + // TargetArch_wasm32: + 3, + // TargetArch_wasm64p32: + 3, +}; + +// Generated with the featuregen script in `misc/featuregen` +gb_global MicroarchFeatureList microarch_features_list[] = { + // TargetArch_amd64: + { str_lit("alderlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("amdfam10"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,vzeroupper,x87") }, + { str_lit("athlon"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-4"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-fx"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-mp"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-tbird"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-xp"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon64"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon64-sse3"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("atom"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,idivl-to-divb,idivq-to-divl,lea-sp,lea-uses-ag,mmx,movbe,no-bypass-delay,nopl,pad-short-functions,sahf,slow-two-mem-ops,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("atom_sse4_2"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("atom_sse4_2_movbe"), str_lit("64bit,64bit-mode,aes,clflushopt,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fsgsbase,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("barcelona"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,vzeroupper,x87") }, + { str_lit("bdver1"), str_lit("64bit,64bit-mode,aes,avx,branchfusion,cmov,crc32,cx16,cx8,fast-11bytenop,fast-scalar-shift-masks,fma4,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,x87,xop,xsave") }, + { str_lit("bdver2"), str_lit("64bit,64bit-mode,aes,avx,bmi,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave") }, + { str_lit("bdver3"), str_lit("64bit,64bit-mode,aes,avx,bmi,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fsgsbase,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave,xsaveopt") }, + { str_lit("bdver4"), str_lit("64bit,64bit-mode,aes,avx,avx2,bmi,bmi2,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fsgsbase,fxsr,lwp,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave,xsaveopt") }, + { str_lit("bonnell"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,idivl-to-divb,idivq-to-divl,lea-sp,lea-uses-ag,mmx,movbe,no-bypass-delay,nopl,pad-short-functions,sahf,slow-two-mem-ops,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("broadwell"), str_lit("64bit,64bit-mode,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("btver1"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fast-15bytenop,fast-scalar-shift-masks,fast-vector-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,ssse3,vzeroupper,x87") }, + { str_lit("btver2"), str_lit("64bit,64bit-mode,aes,avx,bmi,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-hops,fast-lzcnt,fast-movbe,fast-scalar-shift-masks,fast-vector-shift-masks,fxsr,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,x87,xsave,xsaveopt") }, + { str_lit("c3"), str_lit("3dnow,64bit-mode,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("c3-2"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("cannonlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vl,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,sha,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("cascadelake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,avx512vnni,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("cooperlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bf16,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,avx512vnni,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("core-avx-i"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core-avx2"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core2"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("core_2_duo_sse4_1"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,sse4.1,ssse3,vzeroupper,x87") }, + { str_lit("core_2_duo_ssse3"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("core_2nd_gen_avx"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_3rd_gen_avx"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_4th_gen_avx"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_4th_gen_avx_tsx"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_5th_gen_avx"), str_lit("64bit,64bit-mode,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_5th_gen_avx_tsx"), str_lit("64bit,64bit-mode,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_aes_pclmulqdq"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("core_i7_sse4_2"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("corei7"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("corei7-avx"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("emeraldrapids"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("generic"), str_lit("64bit,64bit-mode,cx8,fast-15bytenop,fast-scalar-fsqrt,idivq-to-divl,macrofusion,slow-3ops-lea,sse,sse2,vzeroupper,x87") }, + { str_lit("geode"), str_lit("3dnow,3dnowa,64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("goldmont"), str_lit("64bit,64bit-mode,aes,clflushopt,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("goldmont-plus"), str_lit("64bit,64bit-mode,aes,clflushopt,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("goldmont_plus"), str_lit("64bit,64bit-mode,aes,clflushopt,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("grandridge"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avxifma,avxneconvert,avxvnni,avxvnniint8,bmi,bmi2,cldemote,clflushopt,clwb,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,f16c,fast-movbe,fma,fsgsbase,fxsr,gfni,hreset,invpcid,kl,lzcnt,mmx,movbe,movdir64b,movdiri,no-bypass-delay,nopl,pclmul,pconfig,pku,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,uintr,use-glm-div-sqrt-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids-d"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids_d"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("haswell"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("i386"), str_lit("64bit-mode,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("i486"), str_lit("64bit-mode,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("i586"), str_lit("64bit-mode,cx8,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("i686"), str_lit("64bit-mode,cmov,cx8,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("icelake-client"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake-server"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake_client"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake_server"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("ivybridge"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("k6"), str_lit("64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k6-2"), str_lit("3dnow,64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k6-3"), str_lit("3dnow,64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k8"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k8-sse3"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("knl"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("knm"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,avx512vpopcntdq,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("lakemont"), str_lit("64bit-mode,cx8,slow-unaligned-mem-16,sse,sse2,vzeroupper") }, + { str_lit("meteorlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("mic_avx512"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("nehalem"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("nocona"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("opteron"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("opteron-sse3"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("penryn"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,sse4.1,ssse3,vzeroupper,x87") }, + { str_lit("pentium"), str_lit("64bit-mode,cx8,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium-m"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium-mmx"), str_lit("64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium2"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium3"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium3m"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium4"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium4m"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_4"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_4_sse3"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("pentium_ii"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_iii"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_iii_no_xmm_regs"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_m"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_mmx"), str_lit("64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_pro"), str_lit("64bit-mode,cmov,cx8,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentiumpro"), str_lit("64bit-mode,cmov,cx8,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("prescott"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("raptorlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("rocketlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("sandybridge"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("sapphirerapids"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("sierraforest"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avxifma,avxneconvert,avxvnni,avxvnniint8,bmi,bmi2,cldemote,clflushopt,clwb,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,f16c,fast-movbe,fma,fsgsbase,fxsr,gfni,hreset,invpcid,kl,lzcnt,mmx,movbe,movdir64b,movdiri,no-bypass-delay,nopl,pclmul,pconfig,pku,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,uintr,use-glm-div-sqrt-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("silvermont"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("skx"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake-avx512"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake_avx512"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("slm"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("tigerlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vp2intersect,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("tremont"), str_lit("64bit,64bit-mode,aes,clflushopt,clwb,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,gfni,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("westmere"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("winchip-c6"), str_lit("64bit-mode,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("winchip2"), str_lit("3dnow,64bit-mode,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("x86-64"), str_lit("64bit,64bit-mode,cmov,cx8,fxsr,idivq-to-divl,macrofusion,mmx,nopl,slow-3ops-lea,slow-incdec,sse,sse2,vzeroupper,x87") }, + { str_lit("x86-64-v2"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,nopl,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("x86-64-v3"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fxsr,idivq-to-divl,lzcnt,macrofusion,mmx,movbe,nopl,popcnt,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave") }, + { str_lit("x86-64-v4"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fxsr,idivq-to-divl,lzcnt,macrofusion,mmx,movbe,nopl,popcnt,prefer-256-bit,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave") }, + { str_lit("yonah"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("znver1"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver2"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver3"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,invpcid,lzcnt,macrofusion,mmx,movbe,mwaitx,nopl,pclmul,pku,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver4"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,evex512,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,invpcid,lzcnt,macrofusion,mmx,movbe,mwaitx,nopl,pclmul,pku,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,shstk,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + // TargetArch_i386: + { str_lit("alderlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("amdfam10"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,vzeroupper,x87") }, + { str_lit("athlon"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,mmx,nopl,slow-shld,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("athlon-4"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("athlon-fx"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-mp"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("athlon-tbird"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,mmx,nopl,slow-shld,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("athlon-xp"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("athlon64"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon64-sse3"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("atom"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,idivl-to-divb,idivq-to-divl,lea-sp,lea-uses-ag,mmx,movbe,no-bypass-delay,nopl,pad-short-functions,sahf,slow-two-mem-ops,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("atom_sse4_2"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("atom_sse4_2_movbe"), str_lit("32bit-mode,64bit,aes,clflushopt,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fsgsbase,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("barcelona"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,vzeroupper,x87") }, + { str_lit("bdver1"), str_lit("32bit-mode,64bit,aes,avx,branchfusion,cmov,crc32,cx16,cx8,fast-11bytenop,fast-scalar-shift-masks,fma4,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,x87,xop,xsave") }, + { str_lit("bdver2"), str_lit("32bit-mode,64bit,aes,avx,bmi,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave") }, + { str_lit("bdver3"), str_lit("32bit-mode,64bit,aes,avx,bmi,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fsgsbase,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave,xsaveopt") }, + { str_lit("bdver4"), str_lit("32bit-mode,64bit,aes,avx,avx2,bmi,bmi2,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fsgsbase,fxsr,lwp,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave,xsaveopt") }, + { str_lit("bonnell"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,idivl-to-divb,idivq-to-divl,lea-sp,lea-uses-ag,mmx,movbe,no-bypass-delay,nopl,pad-short-functions,sahf,slow-two-mem-ops,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("broadwell"), str_lit("32bit-mode,64bit,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("btver1"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fast-15bytenop,fast-scalar-shift-masks,fast-vector-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,ssse3,vzeroupper,x87") }, + { str_lit("btver2"), str_lit("32bit-mode,64bit,aes,avx,bmi,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-hops,fast-lzcnt,fast-movbe,fast-scalar-shift-masks,fast-vector-shift-masks,fxsr,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,x87,xsave,xsaveopt") }, + { str_lit("c3"), str_lit("32bit-mode,3dnow,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("c3-2"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("cannonlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vl,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,sha,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("cascadelake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,avx512vnni,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("cooperlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bf16,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,avx512vnni,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("core-avx-i"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core-avx2"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core2"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("core_2_duo_sse4_1"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,sse4.1,ssse3,vzeroupper,x87") }, + { str_lit("core_2_duo_ssse3"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("core_2nd_gen_avx"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_3rd_gen_avx"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_4th_gen_avx"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_4th_gen_avx_tsx"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_5th_gen_avx"), str_lit("32bit-mode,64bit,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_5th_gen_avx_tsx"), str_lit("32bit-mode,64bit,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_aes_pclmulqdq"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("core_i7_sse4_2"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("corei7"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("corei7-avx"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("emeraldrapids"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("generic"), str_lit("32bit-mode,64bit,cx8,fast-15bytenop,fast-scalar-fsqrt,idivq-to-divl,macrofusion,slow-3ops-lea,vzeroupper,x87") }, + { str_lit("geode"), str_lit("32bit-mode,3dnow,3dnowa,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("goldmont"), str_lit("32bit-mode,64bit,aes,clflushopt,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("goldmont-plus"), str_lit("32bit-mode,64bit,aes,clflushopt,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("goldmont_plus"), str_lit("32bit-mode,64bit,aes,clflushopt,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("grandridge"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avxifma,avxneconvert,avxvnni,avxvnniint8,bmi,bmi2,cldemote,clflushopt,clwb,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,f16c,fast-movbe,fma,fsgsbase,fxsr,gfni,hreset,invpcid,kl,lzcnt,mmx,movbe,movdir64b,movdiri,no-bypass-delay,nopl,pclmul,pconfig,pku,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,uintr,use-glm-div-sqrt-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids-d"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids_d"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("haswell"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("i386"), str_lit("32bit-mode,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("i486"), str_lit("32bit-mode,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("i586"), str_lit("32bit-mode,cx8,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("i686"), str_lit("32bit-mode,cmov,cx8,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("icelake-client"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake-server"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake_client"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake_server"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("ivybridge"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("k6"), str_lit("32bit-mode,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("k6-2"), str_lit("32bit-mode,3dnow,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("k6-3"), str_lit("32bit-mode,3dnow,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("k8"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k8-sse3"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("knl"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("knm"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,avx512vpopcntdq,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("lakemont"), str_lit("32bit-mode,cx8,slow-unaligned-mem-16,vzeroupper") }, + { str_lit("meteorlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("mic_avx512"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("nehalem"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("nocona"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("opteron"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("opteron-sse3"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("penryn"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,sse4.1,ssse3,vzeroupper,x87") }, + { str_lit("pentium"), str_lit("32bit-mode,cx8,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium-m"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium-mmx"), str_lit("32bit-mode,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium2"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium3"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("pentium3m"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("pentium4"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium4m"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_4"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_4_sse3"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("pentium_ii"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium_iii"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("pentium_iii_no_xmm_regs"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("pentium_m"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_mmx"), str_lit("32bit-mode,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium_pro"), str_lit("32bit-mode,cmov,cx8,nopl,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentiumpro"), str_lit("32bit-mode,cmov,cx8,nopl,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("prescott"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("raptorlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("rocketlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("sandybridge"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("sapphirerapids"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("sierraforest"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avxifma,avxneconvert,avxvnni,avxvnniint8,bmi,bmi2,cldemote,clflushopt,clwb,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,f16c,fast-movbe,fma,fsgsbase,fxsr,gfni,hreset,invpcid,kl,lzcnt,mmx,movbe,movdir64b,movdiri,no-bypass-delay,nopl,pclmul,pconfig,pku,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,uintr,use-glm-div-sqrt-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("silvermont"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("skx"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake-avx512"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake_avx512"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("slm"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("tigerlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vp2intersect,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("tremont"), str_lit("32bit-mode,64bit,aes,clflushopt,clwb,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,gfni,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("westmere"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("winchip-c6"), str_lit("32bit-mode,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("winchip2"), str_lit("32bit-mode,3dnow,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("x86-64"), str_lit("32bit-mode,64bit,cmov,cx8,fxsr,idivq-to-divl,macrofusion,mmx,nopl,slow-3ops-lea,slow-incdec,sse,sse2,vzeroupper,x87") }, + { str_lit("x86-64-v2"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,nopl,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("x86-64-v3"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fxsr,idivq-to-divl,lzcnt,macrofusion,mmx,movbe,nopl,popcnt,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave") }, + { str_lit("x86-64-v4"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fxsr,idivq-to-divl,lzcnt,macrofusion,mmx,movbe,nopl,popcnt,prefer-256-bit,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave") }, + { str_lit("yonah"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("znver1"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver2"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver3"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,invpcid,lzcnt,macrofusion,mmx,movbe,mwaitx,nopl,pclmul,pku,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver4"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,evex512,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,invpcid,lzcnt,macrofusion,mmx,movbe,mwaitx,nopl,pclmul,pku,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,shstk,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + // TargetArch_arm32: + { str_lit("arm1020e"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm1020t"), str_lit("armv5t,v4t,v5t") }, + { str_lit("arm1022e"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm10e"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm10tdmi"), str_lit("armv5t,v4t,v5t") }, + { str_lit("arm1136j-s"), str_lit("armv6,dsp,v4t,v5t,v5te,v6") }, + { str_lit("arm1136jf-s"), str_lit("armv6,dsp,fp64,fpregs,fpregs64,slowfpvmlx,v4t,v5t,v5te,v6,vfp2,vfp2sp") }, + { str_lit("arm1156t2-s"), str_lit("armv6t2,dsp,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v8m") }, + { str_lit("arm1156t2f-s"), str_lit("armv6t2,dsp,fp64,fpregs,fpregs64,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v8m,vfp2,vfp2sp") }, + { str_lit("arm1176jz-s"), str_lit("armv6kz,trustzone,v4t,v5t,v5te,v6,v6k") }, + { str_lit("arm1176jzf-s"), str_lit("armv6kz,fp64,fpregs,fpregs64,slowfpvmlx,trustzone,v4t,v5t,v5te,v6,v6k,vfp2,vfp2sp") }, + { str_lit("arm710t"), str_lit("armv4t,v4t") }, + { str_lit("arm720t"), str_lit("armv4t,v4t") }, + { str_lit("arm7tdmi"), str_lit("armv4t,v4t") }, + { str_lit("arm7tdmi-s"), str_lit("armv4t,v4t") }, + { str_lit("arm8"), str_lit("armv4") }, + { str_lit("arm810"), str_lit("armv4") }, + { str_lit("arm9"), str_lit("armv4t,v4t") }, + { str_lit("arm920"), str_lit("armv4t,v4t") }, + { str_lit("arm920t"), str_lit("armv4t,v4t") }, + { str_lit("arm922t"), str_lit("armv4t,v4t") }, + { str_lit("arm926ej-s"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm940t"), str_lit("armv4t,v4t") }, + { str_lit("arm946e-s"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm966e-s"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm968e-s"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm9e"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm9tdmi"), str_lit("armv4t,v4t") }, + { str_lit("cortex-a12"), str_lit("a12,aclass,armv7-a,avoid-partial-cpsr,d32,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ret-addr-stack,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vmlx-forwarding") }, + { str_lit("cortex-a15"), str_lit("a15,aclass,armv7-a,avoid-partial-cpsr,d32,db,dont-widen-vmovs,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,muxed-units,neon,perfmon,ret-addr-stack,splat-vfp-neon,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vldn-align") }, + { str_lit("cortex-a17"), str_lit("a17,aclass,armv7-a,avoid-partial-cpsr,d32,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ret-addr-stack,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vmlx-forwarding") }, + { str_lit("cortex-a32"), str_lit("aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a35"), str_lit("a35,aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a5"), str_lit("a5,aclass,armv7-a,d32,db,dsp,fp16,fp64,fpregs,fpregs64,mp,neon,perfmon,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,vmlx-forwarding") }, + { str_lit("cortex-a53"), str_lit("a53,aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpao,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a55"), str_lit("a55,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a57"), str_lit("a57,aclass,acquire-release,aes,armv8-a,avoid-partial-cpsr,cheap-predicable-cpsr,crc,crypto,d32,db,dsp,fix-cortex-a57-aes-1742098,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpao,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a7"), str_lit("a7,aclass,armv7-a,d32,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vmlx-forwarding,vmlx-hazards") }, + { str_lit("cortex-a710"), str_lit("aclass,acquire-release,armv9-a,bf16,cortex-a710,crc,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp16fml,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,i8mm,mp,neon,perfmon,ras,sb,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8m,v9a,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a72"), str_lit("a72,aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fix-cortex-a57-aes-1742098,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a73"), str_lit("a73,aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a75"), str_lit("a75,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a76"), str_lit("a76,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a76ae"), str_lit("a76,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a77"), str_lit("a77,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a78"), str_lit("aclass,acquire-release,aes,armv8.2-a,cortex-a78,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a78c"), str_lit("a78c,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a8"), str_lit("a8,aclass,armv7-a,d32,db,dsp,fp64,fpregs,fpregs64,neon,nonpipelined-vfp,perfmon,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vmlx-forwarding,vmlx-hazards") }, + { str_lit("cortex-a9"), str_lit("a9,aclass,armv7-a,avoid-partial-cpsr,d32,db,dsp,expand-fp-mlx,fp16,fp64,fpregs,fpregs64,mp,muxed-units,neon,neon-fpmovs,perfmon,prefer-vmovsr,ret-addr-stack,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vldn-align,vmlx-forwarding,vmlx-hazards") }, + { str_lit("cortex-m0"), str_lit("armv6-m,db,mclass,no-branch-predictor,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m") }, + { str_lit("cortex-m0plus"), str_lit("armv6-m,db,mclass,no-branch-predictor,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m") }, + { str_lit("cortex-m1"), str_lit("armv6-m,db,mclass,no-branch-predictor,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m") }, + { str_lit("cortex-m23"), str_lit("8msecext,acquire-release,armv8-m.base,db,hwdiv,mclass,no-branch-predictor,no-movt,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m,v7clrex,v8m") }, + { str_lit("cortex-m3"), str_lit("armv7-m,db,hwdiv,loop-align,m3,mclass,no-branch-predictor,noarm,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m") }, + { str_lit("cortex-m33"), str_lit("8msecext,acquire-release,armv8-m.main,db,dsp,fix-cmse-cve-2021-35465,fp-armv8d16sp,fp16,fpregs,hwdiv,loop-align,mclass,no-branch-predictor,noarm,slowfpvfmx,slowfpvmlx,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,v8m.main,vfp2sp,vfp3d16sp,vfp4d16sp") }, + { str_lit("cortex-m35p"), str_lit("8msecext,acquire-release,armv8-m.main,db,dsp,fix-cmse-cve-2021-35465,fp-armv8d16sp,fp16,fpregs,hwdiv,loop-align,mclass,no-branch-predictor,noarm,slowfpvfmx,slowfpvmlx,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,v8m.main,vfp2sp,vfp3d16sp,vfp4d16sp") }, + { str_lit("cortex-m4"), str_lit("armv7e-m,db,dsp,fp16,fpregs,hwdiv,loop-align,mclass,no-branch-predictor,noarm,slowfpvfmx,slowfpvmlx,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2sp,vfp3d16sp,vfp4d16sp") }, + { str_lit("cortex-m55"), str_lit("8msecext,acquire-release,armv8.1-m.main,db,dsp,fix-cmse-cve-2021-35465,fp-armv8d16,fp-armv8d16sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,lob,loop-align,mclass,mve,mve.fp,no-branch-predictor,noarm,ras,slowfpvmlx,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8.1m.main,v8m,v8m.main,vfp2,vfp2sp,vfp3d16,vfp3d16sp,vfp4d16,vfp4d16sp") }, + { str_lit("cortex-m7"), str_lit("armv7e-m,db,dsp,fp-armv8d16,fp-armv8d16sp,fp16,fp64,fpregs,fpregs64,hwdiv,m7,mclass,noarm,thumb-mode,thumb2,use-mipipeliner,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp,vfp4d16,vfp4d16sp") }, + { str_lit("cortex-m85"), str_lit("8msecext,acquire-release,armv8.1-m.main,db,dsp,fp-armv8d16,fp-armv8d16sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,lob,mclass,mve,mve.fp,noarm,pacbti,ras,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8.1m.main,v8m,v8m.main,vfp2,vfp2sp,vfp3d16,vfp3d16sp,vfp4d16,vfp4d16sp") }, + { str_lit("cortex-r4"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,hwdiv,perfmon,r4,rclass,ret-addr-stack,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m") }, + { str_lit("cortex-r4f"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,fp64,fpregs,fpregs64,hwdiv,perfmon,r4,rclass,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp") }, + { str_lit("cortex-r5"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,perfmon,r5,rclass,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp") }, + { str_lit("cortex-r52"), str_lit("acquire-release,armv8-r,crc,d32,db,dfb,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpao,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,r52,rclass,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-r7"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,perfmon,r7,rclass,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp") }, + { str_lit("cortex-r8"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,perfmon,rclass,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp") }, + { str_lit("cortex-x1"), str_lit("aclass,acquire-release,aes,armv8.2-a,cortex-x1,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-x1c"), str_lit("aclass,acquire-release,aes,armv8.2-a,cortex-x1c,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cyclone"), str_lit("aclass,acquire-release,aes,armv8-a,avoid-movs-shop,avoid-partial-cpsr,crc,crypto,d32,db,disable-postra-scheduler,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,neonfp,perfmon,ret-addr-stack,sha2,slowfpvfmx,slowfpvmlx,swift,thumb2,trustzone,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,zcz") }, + { str_lit("ep9312"), str_lit("armv4t,v4t") }, + { str_lit("exynos-m3"), str_lit("aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dont-widen-vmovs,dsp,expand-fp-mlx,exynos,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,fuse-aes,fuse-literals,hwdiv,hwdiv-arm,mp,neon,perfmon,prof-unpr,ret-addr-stack,sha2,slow-fp-brcc,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,splat-vfp-neon,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,wide-stride-vfp,zcz") }, + { str_lit("exynos-m4"), str_lit("aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dont-widen-vmovs,dotprod,dsp,expand-fp-mlx,exynos,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,fuse-aes,fuse-literals,hwdiv,hwdiv-arm,mp,neon,perfmon,prof-unpr,ras,ret-addr-stack,sha2,slow-fp-brcc,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,splat-vfp-neon,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,wide-stride-vfp,zcz") }, + { str_lit("exynos-m5"), str_lit("aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dont-widen-vmovs,dotprod,dsp,expand-fp-mlx,exynos,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,fuse-aes,fuse-literals,hwdiv,hwdiv-arm,mp,neon,perfmon,prof-unpr,ras,ret-addr-stack,sha2,slow-fp-brcc,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,splat-vfp-neon,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,wide-stride-vfp,zcz") }, + { str_lit("generic"), str_lit("") }, + { str_lit("iwmmxt"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("krait"), str_lit("aclass,armv7-a,avoid-partial-cpsr,d32,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,krait,muxed-units,neon,perfmon,ret-addr-stack,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,vldn-align,vmlx-forwarding") }, + { str_lit("kryo"), str_lit("aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,kryo,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("mpcore"), str_lit("armv6k,fp64,fpregs,fpregs64,slowfpvmlx,v4t,v5t,v5te,v6,v6k,vfp2,vfp2sp") }, + { str_lit("mpcorenovfp"), str_lit("armv6k,v4t,v5t,v5te,v6,v6k") }, + { str_lit("neoverse-n1"), str_lit("aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("neoverse-n2"), str_lit("aclass,acquire-release,armv9-a,bf16,crc,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,i8mm,mp,neon,perfmon,ras,sb,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8m,v9a,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("neoverse-v1"), str_lit("aclass,acquire-release,aes,armv8.4-a,bf16,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,i8mm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8.3a,v8.4a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("sc000"), str_lit("armv6-m,db,mclass,no-branch-predictor,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m") }, + { str_lit("sc300"), str_lit("armv7-m,db,hwdiv,m3,mclass,no-branch-predictor,noarm,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m") }, + { str_lit("strongarm"), str_lit("armv4") }, + { str_lit("strongarm110"), str_lit("armv4") }, + { str_lit("strongarm1100"), str_lit("armv4") }, + { str_lit("strongarm1110"), str_lit("armv4") }, + { str_lit("swift"), str_lit("aclass,armv7-a,avoid-movs-shop,avoid-partial-cpsr,d32,db,disable-postra-scheduler,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,neonfp,perfmon,prefer-ishst,prof-unpr,ret-addr-stack,slow-load-D-subreg,slow-odd-reg,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,swift,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,vmlx-hazards,wide-stride-vfp") }, + { str_lit("xscale"), str_lit("armv5te,v4t,v5t,v5te") }, + // TargetArch_arm64: + { str_lit("a64fx"), str_lit("CONTEXTIDREL2,a64fx,aggressive-fma,arith-bcc-fusion,ccpp,complxnum,crc,el2vmsa,el3,fp-armv8,fullfp16,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rdm,sha2,store-pair-suppress,sve,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("ampere1"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,aggressive-fma,altnzcv,alu-lsl-fast,am,ampere1,amvs,arith-bcc-fusion,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,crc,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fptoint,fuse-address,fuse-aes,fuse-literals,i8mm,jsconv,ldp-aligned-only,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,rand,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,stp-aligned-only,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh") }, + { str_lit("ampere1a"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,aggressive-fma,altnzcv,alu-lsl-fast,am,ampere1a,amvs,arith-bcc-fusion,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,crc,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fptoint,fuse-address,fuse-aes,fuse-literals,i8mm,jsconv,ldp-aligned-only,lor,lse,lse2,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predres,rand,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,sm4,specrestrict,ssbs,store-pair-suppress,stp-aligned-only,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh") }, + { str_lit("apple-a10"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a10,arith-bcc-fusion,arith-cbz-fusion,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,lor,neon,pan,perfmon,rdm,sha2,store-pair-suppress,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a11"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a11,arith-bcc-fusion,arith-cbz-fusion,ccpp,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fullfp16,fuse-aes,fuse-crypto-eor,lor,lse,neon,pan,pan-rwv,perfmon,ras,rdm,sha2,store-pair-suppress,uaops,v8.1a,v8.2a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a12"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a12,arith-bcc-fusion,arith-cbz-fusion,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fullfp16,fuse-aes,fuse-crypto-eor,jsconv,lor,lse,neon,pan,pan-rwv,pauth,perfmon,ras,rcpc,rdm,sha2,store-pair-suppress,uaops,v8.1a,v8.2a,v8.3a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a13"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,am,apple-a13,arith-bcc-fusion,arith-cbz-fusion,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,el2vmsa,el3,flagm,fp-armv8,fp16fml,fullfp16,fuse-aes,fuse-crypto-eor,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,ras,rcpc,rcpc-immo,rdm,sel2,sha2,sha3,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a14"), str_lit("CONTEXTIDREL2,aes,aggressive-fma,alternate-sextload-cvt-f32-pattern,altnzcv,am,apple-a14,arith-bcc-fusion,arith-cbz-fusion,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,el2vmsa,el3,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a15"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,altnzcv,am,amvs,apple-a15,arith-bcc-fusion,arith-cbz-fusion,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,i8mm,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a16"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,altnzcv,am,amvs,apple-a16,arith-bcc-fusion,arith-cbz-fusion,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,hcx,i8mm,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a7"), str_lit("aes,alternate-sextload-cvt-f32-pattern,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,neon,perfmon,sha2,store-pair-suppress,v8a,zcm,zcz,zcz-fp-workaround,zcz-gp") }, + { str_lit("apple-a8"), str_lit("aes,alternate-sextload-cvt-f32-pattern,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,neon,perfmon,sha2,store-pair-suppress,v8a,zcm,zcz,zcz-fp-workaround,zcz-gp") }, + { str_lit("apple-a9"), str_lit("aes,alternate-sextload-cvt-f32-pattern,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,neon,perfmon,sha2,store-pair-suppress,v8a,zcm,zcz,zcz-fp-workaround,zcz-gp") }, + { str_lit("apple-latest"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,altnzcv,am,amvs,apple-a16,arith-bcc-fusion,arith-cbz-fusion,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,hcx,i8mm,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-m1"), str_lit("CONTEXTIDREL2,aes,aggressive-fma,alternate-sextload-cvt-f32-pattern,altnzcv,am,apple-a14,arith-bcc-fusion,arith-cbz-fusion,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,el2vmsa,el3,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-m2"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,altnzcv,am,amvs,apple-a15,arith-bcc-fusion,arith-cbz-fusion,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,i8mm,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-s4"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a12,arith-bcc-fusion,arith-cbz-fusion,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fullfp16,fuse-aes,fuse-crypto-eor,jsconv,lor,lse,neon,pan,pan-rwv,pauth,perfmon,ras,rcpc,rdm,sha2,store-pair-suppress,uaops,v8.1a,v8.2a,v8.3a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-s5"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a12,arith-bcc-fusion,arith-cbz-fusion,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fullfp16,fuse-aes,fuse-crypto-eor,jsconv,lor,lse,neon,pan,pan-rwv,pauth,perfmon,ras,rcpc,rdm,sha2,store-pair-suppress,uaops,v8.1a,v8.2a,v8.3a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("carmel"), str_lit("CONTEXTIDREL2,aes,carmel,ccpp,crc,crypto,el2vmsa,el3,fp-armv8,fullfp16,lor,lse,neon,pan,pan-rwv,ras,rdm,sha2,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a34"), str_lit("a35,aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,sha2,v8a") }, + { str_lit("cortex-a35"), str_lit("a35,aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,sha2,v8a") }, + { str_lit("cortex-a510"), str_lit("CONTEXTIDREL2,a510,altnzcv,am,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,dit,dotprod,el2vmsa,el3,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cortex-a53"), str_lit("a53,aes,balance-fp-ops,crc,crypto,el2vmsa,el3,fp-armv8,fuse-adrp-add,fuse-aes,neon,perfmon,sha2,use-postra-scheduler,v8a") }, + { str_lit("cortex-a55"), str_lit("CONTEXTIDREL2,a55,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,ras,rcpc,rdm,sha2,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a57"), str_lit("a57,aes,balance-fp-ops,crc,crypto,el2vmsa,el3,enable-select-opt,fp-armv8,fuse-adrp-add,fuse-aes,fuse-literals,neon,perfmon,predictable-select-expensive,sha2,use-postra-scheduler,v8a") }, + { str_lit("cortex-a65"), str_lit("CONTEXTIDREL2,a65,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-literals,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a65ae"), str_lit("CONTEXTIDREL2,a65,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-literals,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a710"), str_lit("CONTEXTIDREL2,a710,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cortex-a715"), str_lit("CONTEXTIDREL2,a715,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,spe,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cortex-a72"), str_lit("a72,aes,crc,crypto,el2vmsa,el3,enable-select-opt,fp-armv8,fuse-adrp-add,fuse-aes,fuse-literals,neon,perfmon,predictable-select-expensive,sha2,v8a") }, + { str_lit("cortex-a73"), str_lit("a73,aes,crc,crypto,el2vmsa,el3,enable-select-opt,fp-armv8,fuse-adrp-add,fuse-aes,neon,perfmon,predictable-select-expensive,sha2,v8a") }, + { str_lit("cortex-a75"), str_lit("CONTEXTIDREL2,a75,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a76"), str_lit("CONTEXTIDREL2,a76,addr-lsl-fast,aes,alu-lsl-fast,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a76ae"), str_lit("CONTEXTIDREL2,a76,addr-lsl-fast,aes,alu-lsl-fast,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a77"), str_lit("CONTEXTIDREL2,a77,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a78"), str_lit("CONTEXTIDREL2,a78,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a78c"), str_lit("CONTEXTIDREL2,a78c,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,flagm,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-r82"), str_lit("CONTEXTIDREL2,ccidx,ccpp,complxnum,cortex-r82,crc,dit,dotprod,flagm,fp-armv8,fp16fml,fullfp16,jsconv,lse,neon,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,specrestrict,ssbs,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8r") }, + { str_lit("cortex-x1"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,cortex-x1,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-x1c"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,cortex-x1,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,flagm,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,lse2,neon,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,ras,rcpc,rcpc-immo,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-x2"), str_lit("CONTEXTIDREL2,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,cortex-x2,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cortex-x3"), str_lit("CONTEXTIDREL2,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,complxnum,cortex-x3,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,spe,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cyclone"), str_lit("aes,alternate-sextload-cvt-f32-pattern,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,neon,perfmon,sha2,store-pair-suppress,v8a,zcm,zcz,zcz-fp-workaround,zcz-gp") }, + { str_lit("exynos-m3"), str_lit("addr-lsl-fast,aes,alu-lsl-fast,crc,crypto,el2vmsa,el3,exynos-cheap-as-move,exynosm3,force-32bit-jump-tables,fp-armv8,fuse-address,fuse-adrp-add,fuse-aes,fuse-csel,fuse-literals,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,use-postra-scheduler,v8a") }, + { str_lit("exynos-m4"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,arith-bcc-fusion,arith-cbz-fusion,ccpp,crc,crypto,dotprod,el2vmsa,el3,exynos-cheap-as-move,exynosm4,force-32bit-jump-tables,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-csel,fuse-literals,lor,lse,neon,pan,pan-rwv,perfmon,ras,rdm,sha2,store-pair-suppress,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh,zcz,zcz-gp") }, + { str_lit("exynos-m5"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,arith-bcc-fusion,arith-cbz-fusion,ccpp,crc,crypto,dotprod,el2vmsa,el3,exynos-cheap-as-move,exynosm4,force-32bit-jump-tables,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-csel,fuse-literals,lor,lse,neon,pan,pan-rwv,perfmon,ras,rdm,sha2,store-pair-suppress,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh,zcz,zcz-gp") }, + { str_lit("falkor"), str_lit("addr-lsl-fast,aes,alu-lsl-fast,crc,crypto,el2vmsa,el3,falkor,fp-armv8,neon,perfmon,predictable-select-expensive,rdm,sha2,slow-strqro-store,store-pair-suppress,use-postra-scheduler,v8a,zcz,zcz-gp") }, + { str_lit("generic"), str_lit("enable-select-opt,ete,fp-armv8,fuse-adrp-add,fuse-aes,neon,trbe,use-postra-scheduler") }, + { str_lit("kryo"), str_lit("addr-lsl-fast,aes,alu-lsl-fast,crc,crypto,el2vmsa,el3,fp-armv8,kryo,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,use-postra-scheduler,v8a,zcz,zcz-gp") }, + { str_lit("neoverse-512tvb"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,am,bf16,ccdp,ccidx,ccpp,complxnum,crc,crypto,dit,dotprod,el2vmsa,el3,enable-select-opt,flagm,fp-armv8,fp16fml,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mpam,neon,neoverse512tvb,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,rand,ras,rcpc,rcpc-immo,rdm,sel2,sha2,spe,ssbs,sve,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh") }, + { str_lit("neoverse-e1"), str_lit("CONTEXTIDREL2,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,neoversee1,pan,pan-rwv,perfmon,ras,rcpc,rdm,sha2,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("neoverse-n1"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,neoversen1,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("neoverse-n2"), str_lit("CONTEXTIDREL2,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,neoversen2,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("neoverse-v1"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,am,bf16,ccdp,ccidx,ccpp,complxnum,crc,crypto,dit,dotprod,el2vmsa,el3,enable-select-opt,flagm,fp-armv8,fp16fml,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mpam,neon,neoversev1,no-sve-fp-ld1r,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,rand,ras,rcpc,rcpc-immo,rdm,sel2,sha2,spe,ssbs,sve,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh") }, + { str_lit("neoverse-v2"), str_lit("CONTEXTIDREL2,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,neoversev2,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,rand,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,spe,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("saphira"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,am,ccidx,ccpp,complxnum,crc,crypto,dit,dotprod,el2vmsa,el3,flagm,fp-armv8,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,ras,rcpc,rcpc-immo,rdm,saphira,sel2,sha2,spe,store-pair-suppress,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh,zcz,zcz-gp") }, + { str_lit("thunderx"), str_lit("aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,thunderx,use-postra-scheduler,v8a") }, + { str_lit("thunderx2t99"), str_lit("CONTEXTIDREL2,aes,aggressive-fma,arith-bcc-fusion,crc,crypto,el2vmsa,el3,fp-armv8,lor,lse,neon,pan,predictable-select-expensive,rdm,sha2,store-pair-suppress,thunderx2t99,use-postra-scheduler,v8.1a,v8a,vh") }, + { str_lit("thunderx3t110"), str_lit("CONTEXTIDREL2,aes,aggressive-fma,arith-bcc-fusion,balance-fp-ops,ccidx,ccpp,complxnum,crc,crypto,el2vmsa,el3,fp-armv8,jsconv,lor,lse,neon,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,store-pair-suppress,strict-align,thunderx3t110,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8a,vh") }, + { str_lit("thunderxt81"), str_lit("aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,thunderxt81,use-postra-scheduler,v8a") }, + { str_lit("thunderxt83"), str_lit("aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,thunderxt83,use-postra-scheduler,v8a") }, + { str_lit("thunderxt88"), str_lit("aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,thunderxt88,use-postra-scheduler,v8a") }, + { str_lit("tsv110"), str_lit("CONTEXTIDREL2,aes,ccpp,complxnum,crc,crypto,dotprod,el2vmsa,el3,fp-armv8,fp16fml,fullfp16,fuse-aes,jsconv,lor,lse,neon,pan,pan-rwv,perfmon,ras,rdm,sha2,spe,store-pair-suppress,tsv110,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + // TargetArch_wasm32: + { str_lit("bleeding-edge"), str_lit("atomics,bulk-memory,mutable-globals,nontrapping-fptoint,sign-ext,simd128,tail-call") }, + { str_lit("generic"), str_lit("mutable-globals,sign-ext") }, + { str_lit("mvp"), str_lit("") }, + // TargetArch_wasm64p32: + { str_lit("bleeding-edge"), str_lit("atomics,bulk-memory,mutable-globals,nontrapping-fptoint,sign-ext,simd128,tail-call") }, + { str_lit("generic"), str_lit("mutable-globals,sign-ext") }, + { str_lit("mvp"), str_lit("") }, }; gb_global String target_endian_names[TargetEndian_COUNT] = { @@ -164,7 +615,6 @@ struct TargetMetrics { isize max_align; isize max_simd_align; String target_triplet; - String target_data_layout; TargetABIKind abi; }; @@ -445,9 +895,9 @@ struct BuildContext { PtrMap<char const *, ExactValue> defined_values; - BlockingMutex target_features_mutex; StringSet target_features_set; String target_features_string; + bool strict_target_features; String minimum_os_version_string; bool minimum_os_version_string_given; @@ -474,7 +924,18 @@ gb_internal isize MAX_ERROR_COLLECTOR_COUNT(void) { return build_context.max_error_count; } +#if defined(GB_SYSTEM_WINDOWS) + #include <llvm-c/Config/llvm-config.h> +#else + #include <llvm/Config/llvm-config.h> +#endif +// NOTE: AMD64 targets had their alignment on 128 bit ints bumped from 8 to 16 (undocumented of course). +#if LLVM_VERSION_MAJOR >= 18 + #define AMD64_MAX_ALIGNMENT 16 +#else + #define AMD64_MAX_ALIGNMENT 8 +#endif gb_global TargetMetrics target_windows_i386 = { TargetOs_windows, @@ -485,9 +946,8 @@ gb_global TargetMetrics target_windows_i386 = { gb_global TargetMetrics target_windows_amd64 = { TargetOs_windows, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-pc-windows-msvc"), - str_lit("e-m:w-i64:64-f80:128-n8:16:32:64-S128"), }; gb_global TargetMetrics target_linux_i386 = { @@ -500,16 +960,14 @@ gb_global TargetMetrics target_linux_i386 = { gb_global TargetMetrics target_linux_amd64 = { TargetOs_linux, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-pc-linux-gnu"), - str_lit("e-m:w-i64:64-f80:128-n8:16:32:64-S128"), }; gb_global TargetMetrics target_linux_arm64 = { TargetOs_linux, TargetArch_arm64, - 8, 8, 8, 16, + 8, 8, 16, 16, str_lit("aarch64-linux-elf"), - str_lit("e-m:o-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"), }; gb_global TargetMetrics target_linux_arm32 = { @@ -517,15 +975,13 @@ gb_global TargetMetrics target_linux_arm32 = { TargetArch_arm32, 4, 4, 4, 8, str_lit("arm-linux-gnu"), - str_lit("e-m:o-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"), }; gb_global TargetMetrics target_darwin_amd64 = { TargetOs_darwin, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-apple-macosx"), // NOTE: Changes during initialization based on build flags. - str_lit("e-m:o-i64:64-f80:128-n8:16:32:64-S128"), }; gb_global TargetMetrics target_darwin_arm64 = { @@ -533,7 +989,6 @@ gb_global TargetMetrics target_darwin_arm64 = { TargetArch_arm64, 8, 8, 16, 16, str_lit("arm64-apple-macosx"), // NOTE: Changes during initialization based on build flags. - str_lit("e-m:o-i64:64-i128:128-n32:64-S128"), }; gb_global TargetMetrics target_freebsd_i386 = { @@ -546,17 +1001,22 @@ gb_global TargetMetrics target_freebsd_i386 = { gb_global TargetMetrics target_freebsd_amd64 = { TargetOs_freebsd, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-unknown-freebsd-elf"), - str_lit("e-m:w-i64:64-f80:128-n8:16:32:64-S128"), +}; + +gb_global TargetMetrics target_freebsd_arm64 = { + TargetOs_freebsd, + TargetArch_arm64, + 8, 8, 16, 16, + str_lit("aarch64-unknown-freebsd-elf"), }; gb_global TargetMetrics target_openbsd_amd64 = { TargetOs_openbsd, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-unknown-openbsd-elf"), - str_lit("e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"), }; gb_global TargetMetrics target_netbsd_amd64 = { @@ -570,14 +1030,14 @@ gb_global TargetMetrics target_netbsd_amd64 = { gb_global TargetMetrics target_haiku_amd64 = { TargetOs_haiku, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-unknown-haiku"), }; gb_global TargetMetrics target_essence_amd64 = { TargetOs_essence, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-pc-none-elf"), }; @@ -587,7 +1047,6 @@ gb_global TargetMetrics target_freestanding_wasm32 = { TargetArch_wasm32, 4, 4, 8, 16, str_lit("wasm32-freestanding-js"), - str_lit("e-m:e-p:32:32-i64:64-n32:64-S128"), }; gb_global TargetMetrics target_js_wasm32 = { @@ -595,7 +1054,6 @@ gb_global TargetMetrics target_js_wasm32 = { TargetArch_wasm32, 4, 4, 8, 16, str_lit("wasm32-js-js"), - str_lit("e-m:e-p:32:32-i64:64-n32:64-S128"), }; gb_global TargetMetrics target_wasi_wasm32 = { @@ -603,7 +1061,6 @@ gb_global TargetMetrics target_wasi_wasm32 = { TargetArch_wasm32, 4, 4, 8, 16, str_lit("wasm32-wasi-js"), - str_lit("e-m:e-p:32:32-i64:64-n32:64-S128"), }; @@ -612,7 +1069,6 @@ gb_global TargetMetrics target_freestanding_wasm64p32 = { TargetArch_wasm64p32, 4, 8, 8, 16, str_lit("wasm32-freestanding-js"), - str_lit("e-m:e-p:32:32-i64:64-n32:64-S128"), }; gb_global TargetMetrics target_js_wasm64p32 = { @@ -620,7 +1076,6 @@ gb_global TargetMetrics target_js_wasm64p32 = { TargetArch_wasm64p32, 4, 8, 8, 16, str_lit("wasm32-js-js"), - str_lit("e-m:e-p:32:32-i64:64-n32:64-S128"), }; gb_global TargetMetrics target_wasi_wasm64p32 = { @@ -628,7 +1083,6 @@ gb_global TargetMetrics target_wasi_wasm64p32 = { TargetArch_wasm32, 4, 8, 8, 16, str_lit("wasm32-wasi-js"), - str_lit("e-m:e-p:32:32-i64:64-n32:64-S128"), }; @@ -636,27 +1090,24 @@ gb_global TargetMetrics target_wasi_wasm64p32 = { gb_global TargetMetrics target_freestanding_amd64_sysv = { TargetOs_freestanding, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-pc-none-gnu"), - str_lit("e-m:w-i64:64-f80:128-n8:16:32:64-S128"), TargetABI_SysV, }; gb_global TargetMetrics target_freestanding_amd64_win64 = { TargetOs_freestanding, TargetArch_amd64, - 8, 8, 8, 16, + 8, 8, AMD64_MAX_ALIGNMENT, 16, str_lit("x86_64-pc-none-msvc"), - str_lit("e-m:w-i64:64-f80:128-n8:16:32:64-S128"), TargetABI_Win64, }; gb_global TargetMetrics target_freestanding_arm64 = { TargetOs_freestanding, TargetArch_arm64, - 8, 8, 8, 16, + 8, 8, 16, 16, str_lit("aarch64-none-elf"), - str_lit("e-m:o-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"), }; struct NamedTargetMetrics { @@ -680,6 +1131,7 @@ gb_global NamedTargetMetrics named_targets[] = { { str_lit("freebsd_i386"), &target_freebsd_i386 }, { str_lit("freebsd_amd64"), &target_freebsd_amd64 }, + { str_lit("freebsd_arm64"), &target_freebsd_arm64 }, { str_lit("openbsd_amd64"), &target_openbsd_amd64 }, { str_lit("netbsd_amd64"), &target_netbsd_amd64 }, @@ -1435,7 +1887,11 @@ gb_internal void init_build_context(TargetMetrics *cross_target, Subtarget subta metrics = &target_darwin_amd64; #endif #elif defined(GB_SYSTEM_FREEBSD) - metrics = &target_freebsd_amd64; + #if defined(GB_CPU_ARM) + metrics = &target_freebsd_arm64; + #else + metrics = &target_freebsd_amd64; + #endif #elif defined(GB_SYSTEM_OPENBSD) metrics = &target_openbsd_amd64; #elif defined(GB_SYSTEM_NETBSD) @@ -1576,7 +2032,8 @@ gb_internal void init_build_context(TargetMetrics *cross_target, Subtarget subta bc->optimization_level = gb_clamp(bc->optimization_level, -1, 3); - if (bc->metrics.os != TargetOs_windows) { + // TODO: Static map calls are bugged on `amd64sysv` abi. + if (bc->metrics.os != TargetOs_windows && bc->metrics.arch == TargetArch_amd64) { // ENFORCE DYNAMIC MAP CALLS bc->dynamic_map_calls = true; } @@ -1609,48 +2066,53 @@ gb_internal void init_build_context(TargetMetrics *cross_target, Subtarget subta #include "microsoft_craziness.h" #endif +// NOTE: the target feature and microarch lists are all sorted, so if it turns out to be slow (I don't think it will) +// a binary search is possible. -gb_internal Array<String> split_by_comma(String const &list) { - isize n = 1; - for (isize i = 0; i < list.len; i++) { - if (list.text[i] == ',') { - n++; +gb_internal bool check_single_target_feature_is_valid(String const &feature_list, String const &feature) { + String_Iterator it = {feature_list, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (str == feature) { + return true; } } - auto res = array_make<String>(heap_allocator(), n); - String s = list; - for (isize i = 0; i < n; i++) { - isize m = string_index_byte(s, ','); - if (m < 0) { - res[i] = s; - break; + return false; +} + +gb_internal bool check_target_feature_is_valid(String const &feature, TargetArchKind arch, String *invalid) { + String feature_list = target_features_list[arch]; + String_Iterator it = {feature, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!check_single_target_feature_is_valid(feature_list, str)) { + if (invalid) *invalid = str; + return false; } - res[i] = substring(s, 0, m); - s = substring(s, m+1, s.len); } - return res; -} -gb_internal bool check_target_feature_is_valid(TokenPos pos, String const &feature) { - // TODO(bill): check_target_feature_is_valid return true; } -gb_internal bool check_target_feature_is_enabled(TokenPos pos, String const &target_feature_list) { - BuildContext *bc = &build_context; - mutex_lock(&bc->target_features_mutex); - defer (mutex_unlock(&bc->target_features_mutex)); - - auto items = split_by_comma(target_feature_list); - array_free(&items); - for (String const &item : items) { - if (!check_target_feature_is_valid(pos, item)) { - error(pos, "Target feature '%.*s' is not valid", LIT(item)); - return false; +gb_internal bool check_target_feature_is_valid_globally(String const &feature, String *invalid) { + String_Iterator it = {feature, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + + bool valid = false; + for (int arch = TargetArch_Invalid; arch < TargetArch_COUNT; arch += 1) { + if (check_target_feature_is_valid(str, cast(TargetArchKind)arch, invalid)) { + valid = true; + break; + } } - if (!string_set_exists(&bc->target_features_set, item)) { - error(pos, "Target feature '%.*s' is not enabled", LIT(item)); + + if (!valid) { + if (invalid) *invalid = str; return false; } } @@ -1658,54 +2120,35 @@ gb_internal bool check_target_feature_is_enabled(TokenPos pos, String const &tar return true; } -gb_internal void enable_target_feature(TokenPos pos, String const &target_feature_list) { - BuildContext *bc = &build_context; - mutex_lock(&bc->target_features_mutex); - defer (mutex_unlock(&bc->target_features_mutex)); - - auto items = split_by_comma(target_feature_list); - for (String const &item : items) { - if (!check_target_feature_is_valid(pos, item)) { - error(pos, "Target feature '%.*s' is not valid", LIT(item)); - continue; - } - - string_set_add(&bc->target_features_set, item); - } - array_free(&items); +gb_internal bool check_target_feature_is_valid_for_target_arch(String const &feature, String *invalid) { + return check_target_feature_is_valid(feature, build_context.metrics.arch, invalid); } - -gb_internal char const *target_features_set_to_cstring(gbAllocator allocator, bool with_quotes, bool with_plus) { - isize len = 0; - isize i = 0; - for (String const &feature : build_context.target_features_set) { - if (i != 0) { - len += 1; +gb_internal bool check_target_feature_is_enabled(String const &feature, String *not_enabled) { + String_Iterator it = {feature, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!string_set_exists(&build_context.target_features_set, str)) { + if (not_enabled) *not_enabled = str; + return false; } - len += feature.len; - if (with_quotes) len += 2; - if (with_plus) len += 1; - i += 1; } - char *features = gb_alloc_array(allocator, char, len+1); - len = 0; - i = 0; - for (String const &feature : build_context.target_features_set) { - if (i != 0) { - features[len++] = ','; - } - if (with_quotes) features[len++] = '"'; - if (with_plus) features[len++] = '+'; - gb_memmove(features + len, feature.text, feature.len); - len += feature.len; - if (with_quotes) features[len++] = '"'; - i += 1; - } - features[len++] = 0; + return true; +} - return features; +gb_internal bool check_target_feature_is_superset_of(String const &superset, String const &of, String *missing) { + String_Iterator it = {of, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!check_single_target_feature_is_valid(superset, str)) { + if (missing) *missing = str; + return false; + } + } + return true; } // NOTE(Jeroen): Set/create the output and other paths and report an error as appropriate. @@ -1997,10 +2440,6 @@ gb_internal bool init_build_paths(String init_filename) { } } - if (bc->target_features_string.len != 0) { - enable_target_feature({}, bc->target_features_string); - } - return true; } diff --git a/src/check_builtin.cpp b/src/check_builtin.cpp index c3c217ec7..c7d27cf38 100644 --- a/src/check_builtin.cpp +++ b/src/check_builtin.cpp @@ -1719,6 +1719,7 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As case BuiltinProc_objc_register_selector: case BuiltinProc_objc_register_class: case BuiltinProc_atomic_type_is_lock_free: + case BuiltinProc_has_target_feature: // NOTE(bill): The first arg may be a Type, this will be checked case by case break; @@ -2022,6 +2023,7 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As Selection sel = lookup_field(type, field_name, false); if (sel.entity == nullptr) { + ERROR_BLOCK(); gbString type_str = type_to_string_shorthand(type); error(ce->args[0], "'%s' has no field named '%.*s'", type_str, LIT(field_name)); @@ -2095,6 +2097,7 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As Selection sel = lookup_field(type, field_name, false); if (sel.entity == nullptr) { + ERROR_BLOCK(); gbString type_str = type_to_string_shorthand(type); error(ce->args[0], "'%s' has no field named '%.*s'", type_str, LIT(field_name)); @@ -3663,6 +3666,41 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As break; } + case BuiltinProc_has_target_feature: { + String features = str_lit(""); + + check_expr_or_type(c, operand, ce->args[0]); + + if (is_type_string(operand->type) && operand->mode == Addressing_Constant) { + GB_ASSERT(operand->value.kind == ExactValue_String); + features = operand->value.value_string; + } else { + Type *pt = base_type(operand->type); + if (pt->kind == Type_Proc) { + if (pt->Proc.require_target_feature.len != 0) { + GB_ASSERT(pt->Proc.enable_target_feature.len == 0); + features = pt->Proc.require_target_feature; + } else if (pt->Proc.enable_target_feature.len != 0) { + features = pt->Proc.enable_target_feature; + } else { + error(ce->args[0], "Expected the procedure type given to '%.*s' to have @(require_target_feature=\"...\") or @(enable_target_feature=\"...\")", LIT(builtin_name)); + } + } else { + error(ce->args[0], "Expected a constant string or procedure type for '%.*s'", LIT(builtin_name)); + } + } + + String invalid; + if (!check_target_feature_is_valid_globally(features, &invalid)) { + error(ce->args[0], "Target feature '%.*s' is not a valid target feature", LIT(invalid)); + } + + operand->value = exact_value_bool(check_target_feature_is_enabled(features, nullptr)); + operand->mode = Addressing_Constant; + operand->type = t_untyped_bool; + break; + } + case BuiltinProc_soa_struct: { Operand x = {}; Operand y = {}; @@ -5395,6 +5433,58 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As operand->value = exact_value_i64(u->Union.kind == UnionType_no_nil ? 0 : 1); } break; + case BuiltinProc_type_bit_set_elem_type: + { + + if (operand->mode != Addressing_Type) { + error(operand->expr, "Expected a type for '%.*s'", LIT(builtin_name)); + operand->mode = Addressing_Invalid; + operand->type = t_invalid; + return false; + } + + Type *bs = operand->type; + + if (!is_type_bit_set(bs)) { + error(operand->expr, "Expected a bit_set type for '%.*s'", LIT(builtin_name)); + operand->mode = Addressing_Invalid; + operand->type = t_invalid; + return false; + } + + bs = base_type(bs); + GB_ASSERT(bs->kind == Type_BitSet); + + operand->mode = Addressing_Type; + operand->type = bs->BitSet.elem; + } break; + + case BuiltinProc_type_bit_set_underlying_type: + { + + if (operand->mode != Addressing_Type) { + error(operand->expr, "Expected a type for '%.*s'", LIT(builtin_name)); + operand->mode = Addressing_Invalid; + operand->type = t_invalid; + return false; + } + + Type *bs = operand->type; + + if (!is_type_bit_set(bs)) { + error(operand->expr, "Expected a bit_set type for '%.*s'", LIT(builtin_name)); + operand->mode = Addressing_Invalid; + operand->type = t_invalid; + return false; + } + + bs = base_type(bs); + GB_ASSERT(bs->kind == Type_BitSet); + + operand->mode = Addressing_Type; + operand->type = bit_set_to_int(bs); + } break; + case BuiltinProc_type_union_variant_count: { if (operand->mode != Addressing_Type) { @@ -5801,6 +5891,7 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As Selection sel = lookup_field(type, field_name, false); if (sel.entity == nullptr) { + ERROR_BLOCK(); gbString type_str = type_to_string(bt); error(ce->args[0], "'%s' has no field named '%.*s'", type_str, LIT(field_name)); @@ -6014,7 +6105,10 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As return false; } - enable_target_feature({}, str_lit("atomics")); + if (!check_target_feature_is_enabled(str_lit("atomics"), nullptr)) { + error(call, "'%.*s' requires target feature 'atomics' to be enabled, enable it with -target-features:\"atomics\" or choose a different -microarch", LIT(builtin_name)); + return false; + } Operand ptr = {}; Operand expected = {}; @@ -6068,7 +6162,10 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As return false; } - enable_target_feature({}, str_lit("atomics")); + if (!check_target_feature_is_enabled(str_lit("atomics"), nullptr)) { + error(call, "'%.*s' requires target feature 'atomics' to be enabled, enable it with -target-features:\"atomics\" or choose a different -microarch", LIT(builtin_name)); + return false; + } Operand ptr = {}; Operand waiters = {}; diff --git a/src/check_decl.cpp b/src/check_decl.cpp index 952a877a4..441c8000d 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -886,17 +886,37 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { check_objc_methods(ctx, e, ac); - if (ac.require_target_feature.len != 0 && ac.enable_target_feature.len != 0) { - error(e->token, "Attributes @(require_target_feature=...) and @(enable_target_feature=...) cannot be used together"); - } else if (ac.require_target_feature.len != 0) { - if (check_target_feature_is_enabled(e->token.pos, ac.require_target_feature)) { - e->Procedure.target_feature = ac.require_target_feature; - } else { - e->Procedure.target_feature_disabled = true; + { + if (ac.require_target_feature.len != 0 && ac.enable_target_feature.len != 0) { + error(e->token, "A procedure cannot have both @(require_target_feature=\"...\") and @(enable_target_feature=\"...\")"); + } + + if (build_context.strict_target_features && ac.enable_target_feature.len != 0) { + ac.require_target_feature = ac.enable_target_feature; + ac.enable_target_feature.len = 0; + } + + if (ac.require_target_feature.len != 0) { + pt->require_target_feature = ac.require_target_feature; + String invalid; + if (!check_target_feature_is_valid_globally(ac.require_target_feature, &invalid)) { + error(e->token, "Required target feature '%.*s' is not a valid target feature", LIT(invalid)); + } else if (!check_target_feature_is_enabled(ac.require_target_feature, nullptr)) { + e->flags |= EntityFlag_Disabled; + } + } else if (ac.enable_target_feature.len != 0) { + + // NOTE: disallow wasm, features on that arch are always global to the module. + if (is_arch_wasm()) { + error(e->token, "@(enable_target_feature=\"...\") is not allowed on wasm, features for wasm must be declared globally"); + } + + pt->enable_target_feature = ac.enable_target_feature; + String invalid; + if (!check_target_feature_is_valid_globally(ac.enable_target_feature, &invalid)) { + error(e->token, "Procedure enabled target feature '%.*s' is not a valid target feature", LIT(invalid)); + } } - } else if (ac.enable_target_feature.len != 0) { - enable_target_feature(e->token.pos, ac.enable_target_feature); - e->Procedure.target_feature = ac.enable_target_feature; } switch (e->Procedure.optimization_mode) { @@ -1370,6 +1390,10 @@ gb_internal void check_proc_group_decl(CheckerContext *ctx, Entity *pg_entity, D continue; } + if (p->flags & EntityFlag_Disabled) { + continue; + } + String name = p->token.string; for (isize k = j+1; k < pge->entities.count; k++) { @@ -1387,6 +1411,10 @@ gb_internal void check_proc_group_decl(CheckerContext *ctx, Entity *pg_entity, D ERROR_BLOCK(); + if (q->flags & EntityFlag_Disabled) { + continue; + } + ProcTypeOverloadKind kind = are_proc_types_overload_safe(p->type, q->type); bool both_have_where_clauses = false; if (p->decl_info->proc_lit != nullptr && q->decl_info->proc_lit != nullptr) { @@ -1423,6 +1451,7 @@ gb_internal void check_proc_group_decl(CheckerContext *ctx, Entity *pg_entity, D break; case ProcOverload_ParamCount: case ProcOverload_ParamTypes: + case ProcOverload_TargetFeatures: // This is okay :) break; @@ -1590,6 +1619,17 @@ gb_internal bool check_proc_body(CheckerContext *ctx_, Token token, DeclInfo *de if (e->kind != Entity_Variable) { continue; } + if (is_type_polymorphic(e->type)) { + gbString s = type_to_string(e->type); + char const *msg = "Unspecialized polymorphic types are not allowed in procedure parameters, got %s"; + if (e->Variable.type_expr) { + error(e->Variable.type_expr, msg, s); + } else { + error(e->token, msg, s); + } + gb_string_free(s); + } + if (!(e->flags & EntityFlag_Using)) { continue; } diff --git a/src/check_expr.cpp b/src/check_expr.cpp index 06d0a8b12..f0c33d9d8 100644 --- a/src/check_expr.cpp +++ b/src/check_expr.cpp @@ -102,6 +102,7 @@ gb_internal Type * check_init_variable (CheckerContext *c, Entity * gb_internal void check_assignment_error_suggestion(CheckerContext *c, Operand *o, Type *type, i64 max_bit_size=0); gb_internal void add_map_key_type_dependencies(CheckerContext *ctx, Type *key); +gb_internal Type *make_soa_struct_fixed(CheckerContext *ctx, Ast *array_typ_expr, Ast *elem_expr, Type *elem, i64 count, Type *generic_type); gb_internal Type *make_soa_struct_slice(CheckerContext *ctx, Ast *array_typ_expr, Ast *elem_expr, Type *elem); gb_internal Type *make_soa_struct_dynamic_array(CheckerContext *ctx, Ast *array_typ_expr, Ast *elem_expr, Type *elem); @@ -1409,11 +1410,16 @@ gb_internal bool is_polymorphic_type_assignable(CheckerContext *c, Type *poly, T poly->Struct.soa_kind != StructSoa_None) { bool ok = is_polymorphic_type_assignable(c, poly->Struct.soa_elem, source->Struct.soa_elem, true, modify_type); if (ok) switch (source->Struct.soa_kind) { - case StructSoa_Fixed: + case StructSoa_None: default: GB_PANIC("Unhandled SOA Kind"); break; - + case StructSoa_Fixed: + if (modify_type) { + Type *type = make_soa_struct_fixed(c, nullptr, poly->Struct.node, poly->Struct.soa_elem, poly->Struct.soa_count, nullptr); + gb_memmove(poly, type, gb_size_of(*type)); + } + break; case StructSoa_Slice: if (modify_type) { Type *type = make_soa_struct_slice(c, nullptr, poly->Struct.node, poly->Struct.soa_elem); @@ -1802,11 +1808,13 @@ gb_internal bool check_unary_op(CheckerContext *c, Operand *o, Token op) { case Token_Not: if (!is_type_boolean(type) || is_type_array_like(o->type)) { ERROR_BLOCK(); - str = expr_to_string(o->expr); error(op, "Operator '%.*s' is only allowed on boolean expressions", LIT(op.string)); - gb_string_free(str); if (is_type_integer(type)) { - error_line("\tSuggestion: Did you mean to use the bitwise not operator '~'?\n"); + str = expr_to_string(o->expr); + error_line("\tSuggestion: Did you mean to do one of the following?\n"); + error_line("\t\t'%s == 0'?\n", str); + error_line("\t\tUse of the bitwise not operator '~'?\n"); + gb_string_free(str); } } else { o->type = t_untyped_bool; @@ -6185,6 +6193,20 @@ gb_internal bool evaluate_where_clauses(CheckerContext *ctx, Ast *call_expr, Sco } return false; } + + if (ast_file_vet_style(ctx->file)) { + Ast *c = unparen_expr(clause); + if (c->kind == Ast_BinaryExpr && c->BinaryExpr.op.kind == Token_CmpAnd) { + ERROR_BLOCK(); + error(c, "Prefer to separate 'where' clauses with a comma rather than '&&'"); + gbString x = expr_to_string(c->BinaryExpr.left); + gbString y = expr_to_string(c->BinaryExpr.right); + error_line("\tSuggestion: '%s, %s'\n", x, y); + gb_string_free(y); + gb_string_free(x); + } + } + } } @@ -6526,12 +6548,17 @@ gb_internal CallArgumentData check_call_arguments_proc_group(CheckerContext *c, array_add(&proc_entities, proc); } + int max_matched_features = 0; gbString expr_name = expr_to_string(operand->expr); defer (gb_string_free(expr_name)); for_array(i, procs) { Entity *p = procs[i]; + if (p->flags & EntityFlag_Disabled) { + continue; + } + Type *pt = base_type(p->type); if (pt != nullptr && is_type_proc(pt)) { CallArgumentData data = {}; @@ -6562,11 +6589,24 @@ gb_internal CallArgumentData check_call_arguments_proc_group(CheckerContext *c, item.score += assign_score_function(1); } + max_matched_features = gb_max(max_matched_features, matched_target_features(&pt->Proc)); + item.index = index; array_add(&valids, item); } } + if (max_matched_features > 0) { + for_array(i, valids) { + Entity *p = procs[valids[i].index]; + Type *t = base_type(p->type); + GB_ASSERT(t->kind == Type_Proc); + + int matched = matched_target_features(&t->Proc); + valids[i].score += assign_score_function(max_matched_features-matched); + } + } + if (valids.count > 1) { array_sort(valids, valid_index_and_score_cmp); i64 best_score = valids[0].score; @@ -6708,7 +6748,11 @@ gb_internal CallArgumentData check_call_arguments_proc_group(CheckerContext *c, ERROR_BLOCK(); error(operand->expr, "Ambiguous procedure group call '%s' that match with the given arguments", expr_name); - print_argument_types(); + if (positional_operands.count == 0 && named_operands.count == 0) { + error_line("\tNo given arguments\n"); + } else { + print_argument_types(); + } for (auto const &valid : valids) { Entity *proc = proc_entities[valid.index]; @@ -7553,8 +7597,11 @@ gb_internal ExprKind check_call_expr(CheckerContext *c, Operand *operand, Ast *c } } + bool is_call_inlined = false; + switch (inlining) { case ProcInlining_inline: + is_call_inlined = true; if (proc != nullptr) { Entity *e = entity_from_expr(proc); if (e != nullptr && e->kind == Entity_Procedure) { @@ -7570,6 +7617,47 @@ gb_internal ExprKind check_call_expr(CheckerContext *c, Operand *operand, Ast *c break; case ProcInlining_no_inline: break; + case ProcInlining_none: + if (proc != nullptr) { + Entity *e = entity_from_expr(proc); + if (e != nullptr && e->kind == Entity_Procedure) { + DeclInfo *decl = e->decl_info; + if (decl->proc_lit) { + ast_node(pl, ProcLit, decl->proc_lit); + if (pl->inlining == ProcInlining_inline) { + is_call_inlined = true; + } + } + } + } + } + + { + String invalid; + if (pt->kind == Type_Proc && pt->Proc.require_target_feature.len != 0) { + if (!check_target_feature_is_valid_for_target_arch(pt->Proc.require_target_feature, &invalid)) { + error(call, "Called procedure requires target feature '%.*s' which is invalid for the build target", LIT(invalid)); + } else if (!check_target_feature_is_enabled(pt->Proc.require_target_feature, &invalid)) { + error(call, "Calling this procedure requires target feature '%.*s' to be enabled", LIT(invalid)); + } + } + + if (pt->kind == Type_Proc && pt->Proc.enable_target_feature.len != 0) { + if (!check_target_feature_is_valid_for_target_arch(pt->Proc.enable_target_feature, &invalid)) { + error(call, "Called procedure enables target feature '%.*s' which is invalid for the build target", LIT(invalid)); + } + + // NOTE: Due to restrictions in LLVM you can not inline calls with a superset of features. + if (is_call_inlined) { + GB_ASSERT(c->curr_proc_decl); + GB_ASSERT(c->curr_proc_decl->entity); + GB_ASSERT(c->curr_proc_decl->entity->type->kind == Type_Proc); + String scope_features = c->curr_proc_decl->entity->type->Proc.enable_target_feature; + if (!check_target_feature_is_superset_of(scope_features, pt->Proc.enable_target_feature, &invalid)) { + error(call, "Inlined procedure enables target feature '%.*s', this requires the calling procedure to at least enable the same feature", LIT(invalid)); + } + } + } } operand->expr = call; diff --git a/src/check_type.cpp b/src/check_type.cpp index ab8c0b057..11e332757 100644 --- a/src/check_type.cpp +++ b/src/check_type.cpp @@ -381,6 +381,7 @@ gb_internal Type *check_record_polymorphic_params(CheckerContext *ctx, Ast *poly Type *type = nullptr; bool is_type_param = false; bool is_type_polymorphic_type = false; + Type *specialization = nullptr; if (type_expr == nullptr && default_value == nullptr) { error(param, "Expected a type for this parameter"); continue; @@ -393,7 +394,6 @@ gb_internal Type *check_record_polymorphic_params(CheckerContext *ctx, Ast *poly } if (type_expr->kind == Ast_TypeidType) { is_type_param = true; - Type *specialization = nullptr; if (type_expr->TypeidType.specialization != nullptr) { Ast *s = type_expr->TypeidType.specialization; specialization = check_type(ctx, s); @@ -471,6 +471,15 @@ gb_internal Type *check_record_polymorphic_params(CheckerContext *ctx, Ast *poly if (is_type_polymorphic(base_type(operand.type))) { *is_polymorphic_ = true; can_check_fields = false; + } else if (specialization && + !check_type_specialization_to(ctx, specialization, operand.type, false, /*modify_type*/true)) { + if (!ctx->no_polymorphic_errors) { + gbString t = type_to_string(operand.type); + gbString s = type_to_string(specialization); + error(operand.expr, "Cannot convert type '%s' to the specialization '%s'", t, s); + gb_string_free(s); + gb_string_free(t); + } } e = alloc_entity_type_name(scope, token, operand.type); e->TypeName.is_type_alias = true; @@ -797,11 +806,11 @@ gb_internal void check_enum_type(CheckerContext *ctx, Type *enum_type, Type *nam enum_type->Enum.scope = ctx->scope; Type *base_type = t_int; - if (et->base_type != nullptr) { + if (unparen_expr(et->base_type) != nullptr) { base_type = check_type(ctx, et->base_type); } - if (base_type == nullptr || !is_type_integer(base_type)) { + if (base_type == nullptr || base_type == t_invalid || !is_type_integer(base_type)) { error(node, "Base type for enumeration must be an integer"); return; } @@ -1080,6 +1089,8 @@ gb_internal void check_bit_field_type(CheckerContext *ctx, Type *bit_field_type, array_add(&tags, tag); add_entity_use(ctx, field, e); + + total_bit_size += bit_size_u8; } } @@ -1094,7 +1105,7 @@ gb_internal void check_bit_field_type(CheckerContext *ctx, Type *bit_field_type, if (total_bit_size > maximum_bit_size) { gbString s = type_to_string(backing_type); - error(node, "The numbers required %llu exceeds the backing type's (%s) bit size %llu", + error(node, "The total bit size of a bit_field's fields (%llu) must fit into its backing type's (%s) bit size of %llu", cast(unsigned long long)total_bit_size, s, cast(unsigned long long)maximum_bit_size); @@ -1155,7 +1166,7 @@ gb_internal void check_bit_field_type(CheckerContext *ctx, Type *bit_field_type, } } if (all_ones && all_booleans) { - if (build_context.vet_flags & VetFlag_Style) { + if (ast_file_vet_style(ctx->file)) { char const *msg = "This 'bit_field' is better expressed as a 'bit_set' since all of the fields are booleans, of 1-bit in size, and the backing type is an integer (-vet-style)"; error(node, msg); } else { @@ -1428,6 +1439,10 @@ gb_internal bool check_type_specialization_to(CheckerContext *ctx, Type *special bool can_convert = check_cast_internal(ctx, &o, specialization); return can_convert; } else if (t->kind == Type_Struct) { + if (t->Struct.polymorphic_parent == nullptr && + t == s) { + return true; + } if (t->Struct.polymorphic_parent == specialization) { return true; } @@ -1477,6 +1492,10 @@ gb_internal bool check_type_specialization_to(CheckerContext *ctx, Type *special return true; } } else if (t->kind == Type_Union) { + if (t->Union.polymorphic_parent == nullptr && + t == s) { + return true; + } if (t->Union.polymorphic_parent == specialization) { return true; } @@ -2017,8 +2036,8 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para } if (p->flags&FieldFlag_no_alias) { - if (!is_type_pointer(type)) { - error(name, "'#no_alias' can only be applied pointer typed parameters"); + if (!is_type_pointer(type) && !is_type_multi_pointer(type)) { + error(name, "'#no_alias' can only be applied pointer or multi-pointer typed parameters"); p->flags &= ~FieldFlag_no_alias; // Remove the flag } } @@ -2057,6 +2076,7 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para param = alloc_entity_param(scope, name->Ident.token, type, is_using, true); param->Variable.param_value = param_value; param->Variable.field_group_index = field_group_index; + param->Variable.type_expr = type_expr; } } if (p->flags&FieldFlag_no_alias) { @@ -3263,6 +3283,11 @@ gb_internal bool check_type_internal(CheckerContext *ctx, Ast *e, Type **type, T case_end; case_ast_node(pe, ParenExpr, e); + if (pe->expr == nullptr) { + error(e, "Expected an expression or type within the parentheses"); + *type = t_invalid; + return true; + } *type = check_type_expr(ctx, pe->expr, named_type); set_base_type(named_type, *type); return true; diff --git a/src/checker_builtin_procs.hpp b/src/checker_builtin_procs.hpp index c15ec7137..04a8bb848 100644 --- a/src/checker_builtin_procs.hpp +++ b/src/checker_builtin_procs.hpp @@ -44,6 +44,8 @@ enum BuiltinProcId { // "Intrinsics" BuiltinProc_is_package_imported, + BuiltinProc_has_target_feature, + BuiltinProc_transpose, BuiltinProc_outer_product, BuiltinProc_hadamard_product, @@ -267,6 +269,9 @@ BuiltinProc__type_simple_boolean_end, BuiltinProc_type_variant_type_of, BuiltinProc_type_variant_index_of, + BuiltinProc_type_bit_set_elem_type, + BuiltinProc_type_bit_set_underlying_type, + BuiltinProc_type_struct_field_count, BuiltinProc_type_proc_parameter_count, @@ -354,6 +359,8 @@ gb_global BuiltinProc builtin_procs[BuiltinProc_COUNT] = { // "Intrinsics" {STR_LIT("is_package_imported"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, + {STR_LIT("has_target_feature"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, + {STR_LIT("transpose"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, {STR_LIT("outer_product"), 2, false, Expr_Expr, BuiltinProcPkg_intrinsics}, {STR_LIT("hadamard_product"), 2, false, Expr_Expr, BuiltinProcPkg_intrinsics}, @@ -573,6 +580,9 @@ gb_global BuiltinProc builtin_procs[BuiltinProc_COUNT] = { {STR_LIT("type_variant_type_of"), 2, false, Expr_Expr, BuiltinProcPkg_intrinsics}, {STR_LIT("type_variant_index_of"), 2, false, Expr_Expr, BuiltinProcPkg_intrinsics}, + {STR_LIT("type_bit_set_elem_type"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, + {STR_LIT("type_bit_set_underlying_type"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, + {STR_LIT("type_struct_field_count"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, {STR_LIT("type_proc_parameter_count"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, diff --git a/src/entity.cpp b/src/entity.cpp index a12e1d0a6..60ca208ec 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -210,6 +210,7 @@ struct Entity { CommentGroup *comment; } Constant; struct { + Ast *type_expr; // only used for some variables within procedure bodies Ast *init_expr; // only used for some variables within procedure bodies i32 field_index; i32 field_group_index; @@ -252,10 +253,8 @@ struct Entity { bool is_foreign : 1; bool is_export : 1; bool generated_from_polymorphic : 1; - bool target_feature_disabled : 1; bool entry_point_only : 1; bool has_instrumentation : 1; - String target_feature; } Procedure; struct { Array<Entity *> entities; @@ -502,4 +501,4 @@ gb_internal bool is_entity_local_variable(Entity *e) { return ((e->scope->flags &~ ScopeFlag_ContextDefined) == 0) || (e->scope->flags & ScopeFlag_Proc) != 0; -}
\ No newline at end of file +} diff --git a/src/error.cpp b/src/error.cpp index 1b091f88e..688d1b34a 100644 --- a/src/error.cpp +++ b/src/error.cpp @@ -12,7 +12,7 @@ struct ErrorValue { }; struct ErrorCollector { - TokenPos prev; + // TokenPos prev; // no point collecting because of the mulithreaded nature std::atomic<i64> count; std::atomic<i64> warning_count; std::atomic<bool> in_block; @@ -383,14 +383,13 @@ gb_internal void error_va(TokenPos const &pos, TokenPos end, char const *fmt, va } push_error_value(pos, ErrorValue_Error); - // NOTE(bill): Duplicate error, skip it if (pos.line == 0) { error_out_empty(); error_out_coloured("Error: ", TerminalStyle_Normal, TerminalColour_Red); error_out_va(fmt, va); error_out("\n"); - } else if (global_error_collector.prev != pos) { - global_error_collector.prev = pos; + } else { + // global_error_collector.prev = pos; if (json_errors()) { error_out_empty(); } else { @@ -402,10 +401,6 @@ gb_internal void error_va(TokenPos const &pos, TokenPos end, char const *fmt, va error_out_va(fmt, va); error_out("\n"); show_error_on_line(pos, end); - } else { - global_error_collector.curr_error_value = {}; - global_error_collector.curr_error_value_set.store(false); - global_error_collector.count.fetch_sub(1); } try_pop_error_value(); mutex_unlock(&global_error_collector.mutex); @@ -422,14 +417,13 @@ gb_internal void warning_va(TokenPos const &pos, TokenPos end, char const *fmt, push_error_value(pos, ErrorValue_Warning); if (!global_ignore_warnings()) { - // NOTE(bill): Duplicate error, skip it if (pos.line == 0) { error_out_empty(); error_out_coloured("Warning: ", TerminalStyle_Normal, TerminalColour_Yellow); error_out_va(fmt, va); error_out("\n"); - } else if (global_error_collector.prev != pos) { - global_error_collector.prev = pos; + } else { + // global_error_collector.prev = pos; if (json_errors()) { error_out_empty(); } else { @@ -452,21 +446,20 @@ gb_internal void error_line_va(char const *fmt, va_list va) { gb_internal void error_no_newline_va(TokenPos const &pos, char const *fmt, va_list va) { global_error_collector.count.fetch_add(1); + mutex_lock(&global_error_collector.mutex); if (global_error_collector.count.load() > MAX_ERROR_COLLECTOR_COUNT()) { print_all_errors(); gb_exit(1); } - mutex_lock(&global_error_collector.mutex); push_error_value(pos, ErrorValue_Error); - // NOTE(bill): Duplicate error, skip it if (pos.line == 0) { error_out_empty(); error_out_coloured("Error: ", TerminalStyle_Normal, TerminalColour_Red); error_out_va(fmt, va); - } else if (global_error_collector.prev != pos) { - global_error_collector.prev = pos; + } else { + // global_error_collector.prev = pos; if (json_errors()) { error_out_empty(); } else { @@ -485,17 +478,21 @@ gb_internal void error_no_newline_va(TokenPos const &pos, char const *fmt, va_li gb_internal void syntax_error_va(TokenPos const &pos, TokenPos end, char const *fmt, va_list va) { global_error_collector.count.fetch_add(1); + mutex_lock(&global_error_collector.mutex); if (global_error_collector.count > MAX_ERROR_COLLECTOR_COUNT()) { print_all_errors(); gb_exit(1); } - mutex_lock(&global_error_collector.mutex); push_error_value(pos, ErrorValue_Warning); - // NOTE(bill): Duplicate error, skip it - if (global_error_collector.prev != pos) { - global_error_collector.prev = pos; + if (pos.line == 0) { + error_out_empty(); + error_out_coloured("Syntax Error: ", TerminalStyle_Normal, TerminalColour_Red); + error_out_va(fmt, va); + error_out("\n"); + } else { + // global_error_collector.prev = pos; if (json_errors()) { error_out_empty(); } else { @@ -505,11 +502,6 @@ gb_internal void syntax_error_va(TokenPos const &pos, TokenPos end, char const * error_out_va(fmt, va); error_out("\n"); show_error_on_line(pos, end); - } else if (pos.line == 0) { - error_out_empty(); - error_out_coloured("Syntax Error: ", TerminalStyle_Normal, TerminalColour_Red); - error_out_va(fmt, va); - error_out("\n"); } try_pop_error_value(); @@ -518,22 +510,21 @@ gb_internal void syntax_error_va(TokenPos const &pos, TokenPos end, char const * gb_internal void syntax_error_with_verbose_va(TokenPos const &pos, TokenPos end, char const *fmt, va_list va) { global_error_collector.count.fetch_add(1); + mutex_lock(&global_error_collector.mutex); if (global_error_collector.count > MAX_ERROR_COLLECTOR_COUNT()) { print_all_errors(); gb_exit(1); } - mutex_lock(&global_error_collector.mutex); push_error_value(pos, ErrorValue_Warning); - // NOTE(bill): Duplicate error, skip it if (pos.line == 0) { error_out_empty(); error_out_coloured("Syntax_Error: ", TerminalStyle_Normal, TerminalColour_Red); error_out_va(fmt, va); error_out("\n"); - } else if (global_error_collector.prev != pos) { - global_error_collector.prev = pos; + } else { + // global_error_collector.prev = pos; if (json_errors()) { error_out_empty(); } else { @@ -564,9 +555,13 @@ gb_internal void syntax_warning_va(TokenPos const &pos, TokenPos end, char const push_error_value(pos, ErrorValue_Warning); if (!global_ignore_warnings()) { - // NOTE(bill): Duplicate error, skip it - if (global_error_collector.prev != pos) { - global_error_collector.prev = pos; + if (pos.line == 0) { + error_out_empty(); + error_out_coloured("Syntax Warning: ", TerminalStyle_Normal, TerminalColour_Yellow); + error_out_va(fmt, va); + error_out("\n"); + } else { + // global_error_collector.prev = pos; if (json_errors()) { error_out_empty(); } else { @@ -576,11 +571,6 @@ gb_internal void syntax_warning_va(TokenPos const &pos, TokenPos end, char const error_out_va(fmt, va); error_out("\n"); // show_error_on_line(pos, end); - } else if (pos.line == 0) { - error_out_empty(); - error_out_coloured("Syntax Warning: ", TerminalStyle_Normal, TerminalColour_Yellow); - error_out_va(fmt, va); - error_out("\n"); } } @@ -705,9 +695,46 @@ gb_internal void print_all_errors(void) { GB_ASSERT(any_errors() || any_warnings()); - array_sort(global_error_collector.error_values, error_value_cmp); + + { // NOTE(bill): merge neighbouring errors + isize default_lines_to_skip = 1; + if (show_error_line()) { + // NOTE(bill): this will always be 2 extra lines + default_lines_to_skip += 2; + } + + ErrorValue *prev_ev = nullptr; + for (isize i = 0; i < global_error_collector.error_values.count; /**/) { + ErrorValue &ev = global_error_collector.error_values[i]; + + if (prev_ev && prev_ev->pos == ev.pos) { + String_Iterator it = {{ev.msg.data, ev.msg.count}, 0}; + + for (isize lines_to_skip = default_lines_to_skip; lines_to_skip > 0; lines_to_skip -= 1) { + String line = string_split_iterator(&it, '\n'); + if (line.len == 0) { + break; + } + } + + // Merge additional text (suggestions for example) into the previous error. + String current = {prev_ev->msg.data, prev_ev->msg.count}; + String addition = {it.str.text+it.pos, it.str.len-it.pos}; + if (addition.len > 0 && !string_contains_string(current, addition)) { + array_add_elems(&prev_ev->msg, addition.text, addition.len); + } + + array_free(&ev.msg); + array_ordered_remove(&global_error_collector.error_values, i); + } else { + prev_ev = &ev; + i += 1; + } + } + } + gbString res = gb_string_make(heap_allocator(), ""); defer (gb_string_free(res)); @@ -715,6 +742,7 @@ gb_internal void print_all_errors(void) { res = gb_string_append_fmt(res, "{\n"); res = gb_string_append_fmt(res, "\t\"error_count\": %td,\n", global_error_collector.error_values.count); res = gb_string_append_fmt(res, "\t\"errors\": [\n"); + for_array(i, global_error_collector.error_values) { ErrorValue ev = global_error_collector.error_values[i]; @@ -728,9 +756,8 @@ gb_internal void print_all_errors(void) { } res = gb_string_append_fmt(res, "\",\n"); - res = gb_string_append_fmt(res, "\t\t\t\"pos\": {\n"); - if (ev.pos.file_id) { + res = gb_string_append_fmt(res, "\t\t\t\"pos\": {\n"); res = gb_string_append_fmt(res, "\t\t\t\t\"file\": \""); String file = get_file_path_string(ev.pos.file_id); for (isize k = 0; k < file.len; k++) { @@ -743,6 +770,8 @@ gb_internal void print_all_errors(void) { i32 end_column = gb_max(ev.end.column, ev.pos.column); res = gb_string_append_fmt(res, "\t\t\t\t\"end_column\": %d\n", end_column); res = gb_string_append_fmt(res, "\t\t\t},\n"); + } else { + res = gb_string_append_fmt(res, "\t\t\t\"pos\": null,\n"); } res = gb_string_append_fmt(res, "\t\t\t\"msgs\": [\n"); diff --git a/src/exact_value.cpp b/src/exact_value.cpp index 83af82f55..1a42a82a9 100644 --- a/src/exact_value.cpp +++ b/src/exact_value.cpp @@ -121,12 +121,14 @@ gb_internal ExactValue exact_value_string(String string) { gb_internal ExactValue exact_value_i64(i64 i) { ExactValue result = {ExactValue_Integer}; + result.value_integer = {0}; big_int_from_i64(&result.value_integer, i); return result; } gb_internal ExactValue exact_value_u64(u64 i) { ExactValue result = {ExactValue_Integer}; + result.value_integer = {0}; big_int_from_u64(&result.value_integer, i); return result; } @@ -177,6 +179,7 @@ gb_internal ExactValue exact_value_typeid(Type *type) { gb_internal ExactValue exact_value_integer_from_string(String const &string) { ExactValue result = {ExactValue_Integer}; + result.value_integer = {0}; bool success; big_int_from_string(&result.value_integer, string, &success); if (!success) { @@ -585,6 +588,7 @@ gb_internal ExactValue exact_unary_operator_value(TokenKind op, ExactValue v, i3 return v; case ExactValue_Integer: { ExactValue i = {ExactValue_Integer}; + i.value_integer = {0}; big_int_neg(&i.value_integer, &v.value_integer); return i; } @@ -616,6 +620,7 @@ gb_internal ExactValue exact_unary_operator_value(TokenKind op, ExactValue v, i3 case ExactValue_Integer: { GB_ASSERT(precision != 0); ExactValue i = {ExactValue_Integer}; + i.value_integer = {0}; big_int_not(&i.value_integer, &v.value_integer, precision, !is_unsigned); return i; } diff --git a/src/gb/gb.h b/src/gb/gb.h index 4e05a3a0a..58fbfa071 100644 --- a/src/gb/gb.h +++ b/src/gb/gb.h @@ -3025,6 +3025,8 @@ gb_inline u32 gb_thread_current_id(void) { thread_id = gettid(); #elif defined(GB_SYSTEM_HAIKU) thread_id = find_thread(NULL); +#elif defined(GB_SYSTEM_FREEBSD) + thread_id = pthread_getthreadid_np(); #else #error Unsupported architecture for gb_thread_current_id() #endif diff --git a/src/linker.cpp b/src/linker.cpp index e694fd999..33ca70462 100644 --- a/src/linker.cpp +++ b/src/linker.cpp @@ -432,7 +432,7 @@ gb_internal i32 linker_stage(LinkerData *gen) { if (string_ends_with(lib, str_lit(".a")) || string_ends_with(lib, str_lit(".o"))) { // static libs and object files, absolute full path relative to the file in which the lib was imported from lib_str = gb_string_append_fmt(lib_str, " -l:\"%.*s\" ", LIT(lib)); - } else if (string_ends_with(lib, str_lit(".so"))) { + } else if (string_ends_with(lib, str_lit(".so")) || string_contains_string(lib, str_lit(".so."))) { // dynamic lib, relative path to executable // NOTE(vassvik): it is the user's responsibility to make sure the shared library files are visible // at runtime to the executable diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index 4b94cf020..03c17a8bb 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -41,6 +41,37 @@ String get_default_microarchitecture() { return default_march; } +String get_final_microarchitecture() { + BuildContext *bc = &build_context; + + String microarch = bc->microarch; + if (microarch.len == 0) { + microarch = get_default_microarchitecture(); + } else if (microarch == str_lit("native")) { + microarch = make_string_c(LLVMGetHostCPUName()); + } + return microarch; +} + +gb_internal String get_default_features() { + BuildContext *bc = &build_context; + + int off = 0; + for (int i = 0; i < bc->metrics.arch; i += 1) { + off += target_microarch_counts[i]; + } + + String microarch = get_final_microarchitecture(); + for (int i = off; i < off+target_microarch_counts[bc->metrics.arch]; i += 1) { + if (microarch_features_list[i].microarch == microarch) { + return microarch_features_list[i].features; + } + } + + GB_PANIC("unknown microarch"); + return {}; +} + gb_internal void lb_add_foreign_library_path(lbModule *m, Entity *e) { if (e == nullptr) { return; @@ -1477,6 +1508,7 @@ gb_internal WORKER_TASK_PROC(lb_llvm_module_pass_worker_proc) { case 1: // default<Os> // Passes removed: coro, openmp, sroa +#if LLVM_VERSION_MAJOR == 17 array_add(&passes, u8R"( annotation2metadata, forceattrs, @@ -1492,13 +1524,14 @@ globalopt, function<eager-inv>( mem2reg, instcombine<max-iterations=1000;no-use-loop-info>, - simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>), - require<globals-aa>, - function( - invalidate<aa> - ), - require<profile-summary>, - cgscc( + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch> +), +require<globals-aa>, +function( + invalidate<aa> +), +require<profile-summary>, +cgscc( devirt<4>( inline<only-mandatory>, inline, @@ -1599,10 +1632,142 @@ function( ), verify )"); +#else + array_add(&passes, u8R"( +annotation2metadata, +forceattrs, +inferattrs, +function<eager-inv>( + lower-expect, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;no-switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + sroa<modify-cfg>, + early-cse<> +), +ipsccp, +called-value-propagation, +globalopt, +function<eager-inv>( + mem2reg, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch> +), +always-inline, +require<globals-aa>, +function( + invalidate<aa> +), +require<profile-summary>, +cgscc( + devirt<4>( + inline, + function-attrs<skip-non-recursive-function-attrs>, + function<eager-inv;no-rerun>( + sroa<modify-cfg>, + early-cse<memssa>, + speculative-execution<only-if-divergent-target>, + jump-threading, + correlated-propagation, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + aggressive-instcombine, + tailcallelim, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + reassociate, + constraint-elimination, + loop-mssa( + loop-instsimplify, + loop-simplifycfg, + licm<no-allowspeculation>, + loop-rotate<header-duplication;no-prepare-for-lto>, + licm<allowspeculation>, + simple-loop-unswitch<no-nontrivial;trivial> + ), + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop( + loop-idiom, + indvars, + loop-deletion, + loop-unroll-full + ), + sroa<modify-cfg>, + vector-combine, + mldst-motion<no-split-footer-bb>, + gvn<>, + sccp, + bdce, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + jump-threading, + correlated-propagation, + adce, + memcpyopt, + dse, + move-auto-init, + loop-mssa( + licm<allowspeculation> + ), + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;hoist-common-insts;sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint> + ), + function-attrs, + function( + require<should-not-run-function-passes> + ) + ) +), +deadargelim, +globalopt, +globaldce, +elim-avail-extern, +rpo-function-attrs, +recompute-globalsaa, +function<eager-inv>( + float2int, + lower-constant-intrinsics, + loop( + loop-rotate<header-duplication;no-prepare-for-lto>, + loop-deletion + ), + loop-distribute, + inject-tli-mappings, + loop-vectorize<no-interleave-forced-only;no-vectorize-forced-only;>, + infer-alignment, + loop-load-elim, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + simplifycfg<bonus-inst-threshold=1;forward-switch-cond;switch-range-to-icmp;switch-to-lookup;no-keep-loops;hoist-common-insts;sink-common-insts;speculate-blocks;simplify-cond-branch>, + slp-vectorizer, + vector-combine, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop-unroll<O2>, + transform-warning, + sroa<preserve-cfg>, + infer-alignment, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop-mssa( + licm<allowspeculation> + ), + alignment-from-assumptions, + loop-sink, + instsimplify, + div-rem-pairs, + tailcallelim, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch> +), +globaldce, +constmerge, +cg-profile, +rel-lookup-table-converter, +function( + annotation-remarks +), +verify +)"); +#endif break; // default<O2> // Passes removed: coro, openmp, sroa case 2: +#if LLVM_VERSION_MAJOR == 17 array_add(&passes, u8R"( annotation2metadata, forceattrs, @@ -1727,11 +1892,144 @@ function( ), verify )"); +#else + array_add(&passes, u8R"( +annotation2metadata, +forceattrs, +inferattrs, +function<eager-inv>( + lower-expect, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;no-switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + sroa<modify-cfg>, + early-cse<> +), +ipsccp, +called-value-propagation, +globalopt, +function<eager-inv>( + mem2reg, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch> +), +always-inline, +require<globals-aa>, +function( + invalidate<aa> +), +require<profile-summary>, +cgscc( + devirt<4>( + inline, + function-attrs<skip-non-recursive-function-attrs>, + function<eager-inv;no-rerun>( + sroa<modify-cfg>, + early-cse<memssa>, + speculative-execution<only-if-divergent-target>, + jump-threading, + correlated-propagation, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + aggressive-instcombine, + libcalls-shrinkwrap, + tailcallelim, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + reassociate, + constraint-elimination, + loop-mssa( + loop-instsimplify, + loop-simplifycfg, + licm<no-allowspeculation>, + loop-rotate<header-duplication;no-prepare-for-lto>, + licm<allowspeculation>, + simple-loop-unswitch<no-nontrivial;trivial> + ), + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop( + loop-idiom, + indvars, + loop-deletion, + loop-unroll-full + ), + sroa<modify-cfg>, + vector-combine, + mldst-motion<no-split-footer-bb>, + gvn<>, + sccp, + bdce, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + jump-threading, + correlated-propagation, + adce, + memcpyopt, + dse, + move-auto-init, + loop-mssa( + licm<allowspeculation> + ), + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;hoist-common-insts;sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint> + ), + function-attrs, + function( + require<should-not-run-function-passes> + ) + ) +), +deadargelim, +globalopt, +globaldce, +elim-avail-extern, +rpo-function-attrs, +recompute-globalsaa, +function<eager-inv>( + float2int, + lower-constant-intrinsics, + loop( + loop-rotate<header-duplication;no-prepare-for-lto>, + loop-deletion + ), + loop-distribute, + inject-tli-mappings, + loop-vectorize<no-interleave-forced-only;no-vectorize-forced-only;>, + infer-alignment, + loop-load-elim, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + simplifycfg<bonus-inst-threshold=1;forward-switch-cond;switch-range-to-icmp;switch-to-lookup;no-keep-loops;hoist-common-insts;sink-common-insts;speculate-blocks;simplify-cond-branch>, + slp-vectorizer, + vector-combine, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop-unroll<O2>, + transform-warning, + sroa<modify-cfg>, + infer-alignment, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop-mssa( + licm<allowspeculation> + ), + alignment-from-assumptions, + loop-sink, + instsimplify, + div-rem-pairs, + tailcallelim, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch> +), +globaldce, +constmerge, +cg-profile, +rel-lookup-table-converter, +function( + annotation-remarks +), +verify +)"); +#endif break; case 3: // default<O3> // Passes removed: coro, openmp, sroa +#if LLVM_VERSION_MAJOR == 17 array_add(&passes, u8R"( annotation2metadata, forceattrs, @@ -1859,6 +2157,135 @@ function( ), verify )"); +#else + array_add(&passes, u8R"( +annotation2metadata, +forceattrs, +inferattrs, +function<eager-inv>( + lower-expect, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;no-switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + sroa<modify-cfg>, + early-cse<>, + callsite-splitting +), +ipsccp, +called-value-propagation, +globalopt, +function<eager-inv>( + mem2reg, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch> +), +always-inline, +require<globals-aa>, +function(invalidate<aa>), +require<profile-summary>, +cgscc( + devirt<4>( + inline, + function-attrs<skip-non-recursive-function-attrs>, + argpromotion, + function<eager-inv;no-rerun>( + sroa<modify-cfg>, + early-cse<memssa>, + speculative-execution<only-if-divergent-target>, + jump-threading, + correlated-propagation, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + aggressive-instcombine, + libcalls-shrinkwrap, + tailcallelim, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + reassociate, + constraint-elimination, + loop-mssa( + loop-instsimplify, + loop-simplifycfg, + licm<no-allowspeculation>, + loop-rotate<header-duplication;no-prepare-for-lto>, + licm<allowspeculation>, + simple-loop-unswitch<nontrivial;trivial> + ), + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop( + loop-idiom, + indvars, + loop-deletion, + loop-unroll-full + ), + sroa<modify-cfg>, + vector-combine, + mldst-motion<no-split-footer-bb>, + gvn<>, + sccp, + bdce, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + jump-threading, + correlated-propagation, + adce, + memcpyopt, + dse, + move-auto-init, + loop-mssa(licm<allowspeculation>), + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;hoist-common-insts;sink-common-insts;speculate-blocks;simplify-cond-branch>, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint> + ), + function-attrs, + function( + require<should-not-run-function-passes> + ) + ) +), +deadargelim, +globalopt, +globaldce, +elim-avail-extern, +rpo-function-attrs, +recompute-globalsaa, +function<eager-inv>( + float2int, + lower-constant-intrinsics, + chr, + loop( + loop-rotate<header-duplication;no-prepare-for-lto>, + loop-deletion + ), + loop-distribute, + inject-tli-mappings, + loop-vectorize<no-interleave-forced-only;no-vectorize-forced-only;>, + infer-alignment, + loop-load-elim, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + simplifycfg<bonus-inst-threshold=1;forward-switch-cond;switch-range-to-icmp;switch-to-lookup;no-keep-loops;hoist-common-insts;sink-common-insts;speculate-blocks;simplify-cond-branch>, + slp-vectorizer, + vector-combine, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop-unroll<O3>, + transform-warning, + sroa<preserve-cfg>, + infer-alignment, + instcombine<max-iterations=1;no-use-loop-info;no-verify-fixpoint>, + loop-mssa(licm<allowspeculation>), + alignment-from-assumptions, + loop-sink, + instsimplify, + div-rem-pairs, + tailcallelim, + simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts;speculate-blocks;simplify-cond-branch> +), +globaldce, +constmerge, +cg-profile, +rel-lookup-table-converter, +function( + annotation-remarks +), +verify +)"); +#endif break; } @@ -2468,69 +2895,24 @@ gb_internal bool lb_generate_code(lbGenerator *gen) { code_mode = LLVMCodeModelKernel; } - String host_cpu_name = copy_string(permanent_allocator(), make_string_c(LLVMGetHostCPUName())); - String llvm_cpu = get_default_microarchitecture(); - char const *llvm_features = ""; - if (build_context.microarch.len != 0) { - if (build_context.microarch == "native") { - llvm_cpu = host_cpu_name; - } else { - llvm_cpu = copy_string(permanent_allocator(), build_context.microarch); - } - if (llvm_cpu == host_cpu_name) { - llvm_features = LLVMGetHostCPUFeatures(); + String llvm_cpu = get_final_microarchitecture(); + + gbString llvm_features = gb_string_make(temporary_allocator(), ""); + String_Iterator it = {build_context.target_features_string, 0}; + bool first = true; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!first) { + llvm_features = gb_string_appendc(llvm_features, ","); } - } + first = false; - // NOTE(Jeroen): Uncomment to get the list of supported microarchitectures. - /* - if (build_context.microarch == "?") { - string_set_add(&build_context.target_features_set, str_lit("+cpuhelp")); + llvm_features = gb_string_appendc(llvm_features, "+"); + llvm_features = gb_string_append_length(llvm_features, str.text, str.len); } - */ - if (build_context.target_features_set.entries.count != 0) { - // Prefix all of the features with a `+`, because we are - // enabling additional features. - char const *additional_features = target_features_set_to_cstring(permanent_allocator(), false, true); - - String f_string = make_string_c(llvm_features); - String a_string = make_string_c(additional_features); - isize f_len = f_string.len; - - if (f_len == 0) { - // The common case is that llvm_features is empty, so - // the target_features_set additions can be used as is. - llvm_features = additional_features; - } else { - // The user probably specified `-microarch:native`, so - // llvm_features is populated by LLVM's idea of what - // the host CPU supports. - // - // As far as I can tell, (which is barely better than - // wild guessing), a bitset is formed by parsing the - // string left to right. - // - // So, llvm_features + ',' + additonal_features, will - // makes the target_features_set override llvm_features. - - char *tmp = gb_alloc_array(permanent_allocator(), char, f_len + 1 + a_string.len + 1); - isize len = 0; - - // tmp = f_string - gb_memmove(tmp, f_string.text, f_string.len); - len += f_string.len; - // tmp += ',' - tmp[len++] = ','; - // tmp += a_string - gb_memmove(tmp + len, a_string.text, a_string.len); - len += a_string.len; - // tmp += NUL - tmp[len++] = 0; - - llvm_features = tmp; - } - } + debugf("CPU: %.*s, Features: %s\n", LIT(llvm_cpu), llvm_features); // GB_ASSERT_MSG(LLVMTargetHasAsmBackend(target)); diff --git a/src/llvm_backend.hpp b/src/llvm_backend.hpp index 7dc5f6b63..9f7bc8843 100644 --- a/src/llvm_backend.hpp +++ b/src/llvm_backend.hpp @@ -1,13 +1,9 @@ #if defined(GB_SYSTEM_WINDOWS) -#include "llvm-c/Core.h" -#include "llvm-c/ExecutionEngine.h" -#include "llvm-c/Target.h" -#include "llvm-c/Analysis.h" -#include "llvm-c/Object.h" -#include "llvm-c/BitWriter.h" -#include "llvm-c/DebugInfo.h" -#include "llvm-c/Transforms/PassBuilder.h" +#include <llvm-c/Config/llvm-config.h> #else +#include <llvm/Config/llvm-config.h> +#endif + #include <llvm-c/Core.h> #include <llvm-c/ExecutionEngine.h> #include <llvm-c/Target.h> @@ -26,7 +22,6 @@ #include <llvm-c/Transforms/Utils.h> #include <llvm-c/Transforms/Vectorize.h> #endif -#endif #if LLVM_VERSION_MAJOR < 11 #error "LLVM Version 11 is the minimum required" @@ -581,6 +576,10 @@ gb_internal LLVMTypeRef llvm_array_type(LLVMTypeRef ElementType, uint64_t Elemen #endif } + +gb_internal void lb_set_metadata_custom_u64(lbModule *m, LLVMValueRef v_ref, String name, u64 value); +gb_internal u64 lb_get_metadata_custom_u64(lbModule *m, LLVMValueRef v_ref, String name); + #define LB_STARTUP_RUNTIME_PROC_NAME "__$startup_runtime" #define LB_CLEANUP_RUNTIME_PROC_NAME "__$cleanup_runtime" #define LB_TYPE_INFO_DATA_NAME "__$type_info_data" @@ -719,4 +718,4 @@ gb_global char const *llvm_linkage_strings[] = { "linker private weak linkage" }; -#define ODIN_METADATA_REQUIRE "odin-metadata-require", 21 +#define ODIN_METADATA_IS_PACKED str_lit("odin-is-packed")
\ No newline at end of file diff --git a/src/llvm_backend_expr.cpp b/src/llvm_backend_expr.cpp index ee1a384ae..030926079 100644 --- a/src/llvm_backend_expr.cpp +++ b/src/llvm_backend_expr.cpp @@ -4383,7 +4383,11 @@ gb_internal lbAddr lb_build_addr_compound_lit(lbProcedure *p, Ast *expr) { mask = LLVMConstSub(mask, LLVMConstInt(lit, 1, false)); LLVMValueRef elem = values[i].value; - elem = LLVMBuildZExt(p->builder, elem, lit, ""); + if (lb_sizeof(lit) < lb_sizeof(LLVMTypeOf(elem))) { + elem = LLVMBuildTrunc(p->builder, elem, lit, ""); + } else { + elem = LLVMBuildZExt(p->builder, elem, lit, ""); + } elem = LLVMBuildAnd(p->builder, elem, mask, ""); elem = LLVMBuildShl(p->builder, elem, LLVMConstInt(lit, f.bit_offset, false), ""); diff --git a/src/llvm_backend_general.cpp b/src/llvm_backend_general.cpp index 15cbb7c71..9caddfb51 100644 --- a/src/llvm_backend_general.cpp +++ b/src/llvm_backend_general.cpp @@ -1107,6 +1107,12 @@ gb_internal lbValue lb_emit_load(lbProcedure *p, lbValue value) { GB_ASSERT(is_type_pointer(value.type)); Type *t = type_deref(value.type); LLVMValueRef v = LLVMBuildLoad2(p->builder, lb_type(p->module, t), value.value, ""); + + u64 is_packed = lb_get_metadata_custom_u64(p->module, value.value, ODIN_METADATA_IS_PACKED); + if (is_packed != 0) { + LLVMSetAlignment(v, 1); + } + return lbValue{v, t}; } diff --git a/src/llvm_backend_proc.cpp b/src/llvm_backend_proc.cpp index f73698d34..898c9ac31 100644 --- a/src/llvm_backend_proc.cpp +++ b/src/llvm_backend_proc.cpp @@ -177,17 +177,24 @@ gb_internal lbProcedure *lb_create_procedure(lbModule *m, Entity *entity, bool i break; } - if (!entity->Procedure.target_feature_disabled && - entity->Procedure.target_feature.len != 0) { - auto features = split_by_comma(entity->Procedure.target_feature); - for_array(i, features) { - String feature = features[i]; - LLVMAttributeRef ref = LLVMCreateStringAttribute( - m->ctx, - cast(char const *)feature.text, cast(unsigned)feature.len, - "", 0); - LLVMAddAttributeAtIndex(p->value, LLVMAttributeIndex_FunctionIndex, ref); + if (pt->Proc.enable_target_feature.len != 0) { + gbString feature_str = gb_string_make(temporary_allocator(), ""); + + String_Iterator it = {pt->Proc.enable_target_feature, 0}; + bool first = true; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!first) { + feature_str = gb_string_appendc(feature_str, ","); + } + first = false; + + feature_str = gb_string_appendc(feature_str, "+"); + feature_str = gb_string_append_length(feature_str, str.text, str.len); } + + lb_add_attribute_to_proc_with_string(m, p->value, make_string_c("target-features"), make_string_c(feature_str)); } if (entity->flags & EntityFlag_Cold) { diff --git a/src/llvm_backend_utility.cpp b/src/llvm_backend_utility.cpp index c01ab0692..8b046c784 100644 --- a/src/llvm_backend_utility.cpp +++ b/src/llvm_backend_utility.cpp @@ -57,6 +57,29 @@ gb_internal lbValue lb_correct_endianness(lbProcedure *p, lbValue value) { return value; } + +gb_internal void lb_set_metadata_custom_u64(lbModule *m, LLVMValueRef v_ref, String name, u64 value) { + unsigned md_id = LLVMGetMDKindIDInContext(m->ctx, cast(char const *)name.text, cast(unsigned)name.len); + LLVMMetadataRef md = LLVMValueAsMetadata(LLVMConstInt(lb_type(m, t_u64), value, false)); + LLVMValueRef node = LLVMMetadataAsValue(m->ctx, LLVMMDNodeInContext2(m->ctx, &md, 1)); + LLVMSetMetadata(v_ref, md_id, node); +} +gb_internal u64 lb_get_metadata_custom_u64(lbModule *m, LLVMValueRef v_ref, String name) { + unsigned md_id = LLVMGetMDKindIDInContext(m->ctx, cast(char const *)name.text, cast(unsigned)name.len); + LLVMValueRef v_md = LLVMGetMetadata(v_ref, md_id); + if (v_md == nullptr) { + return 0; + } + unsigned node_count = LLVMGetMDNodeNumOperands(v_md); + if (node_count == 0) { + return 0; + } + GB_ASSERT(node_count == 1); + LLVMValueRef value = nullptr; + LLVMGetMDNodeOperands(v_md, &value); + return LLVMConstIntGetZExtValue(value); +} + gb_internal LLVMValueRef lb_mem_zero_ptr_internal(lbProcedure *p, LLVMValueRef ptr, usize len, unsigned alignment, bool is_volatile) { return lb_mem_zero_ptr_internal(p, ptr, LLVMConstInt(lb_type(p->module, t_uint), len, false), alignment, is_volatile); } @@ -1148,7 +1171,15 @@ gb_internal lbValue lb_emit_struct_ep(lbProcedure *p, lbValue s, i32 index) { GB_ASSERT_MSG(result_type != nullptr, "%s %d", type_to_string(t), index); - return lb_emit_struct_ep_internal(p, s, index, result_type); + lbValue gep = lb_emit_struct_ep_internal(p, s, index, result_type); + + Type *bt = base_type(t); + if (bt->kind == Type_Struct && bt->Struct.is_packed) { + lb_set_metadata_custom_u64(p->module, gep.value, ODIN_METADATA_IS_PACKED, 1); + GB_ASSERT(lb_get_metadata_custom_u64(p->module, gep.value, ODIN_METADATA_IS_PACKED) == 1); + } + + return gep; } gb_internal lbValue lb_emit_tuple_ev(lbProcedure *p, lbValue value, i32 index) { @@ -1708,7 +1739,8 @@ gb_internal lbValue lb_emit_mul_add(lbProcedure *p, lbValue a, lbValue b, lbValu if (is_possible) { switch (build_context.metrics.arch) { case TargetArch_amd64: - if (type_size_of(t) == 2) { + // NOTE: using the intrinsic when not supported causes slow codegen (See #2928). + if (type_size_of(t) == 2 || !check_target_feature_is_enabled(str_lit("fma"), nullptr)) { is_possible = false; } break; diff --git a/src/main.cpp b/src/main.cpp index ee7de7f81..e76b0e380 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -86,11 +86,8 @@ gb_global Timings global_timings = {0}; #if defined(GB_SYSTEM_OSX) #include <llvm/Config/llvm-config.h> - #if LLVM_VERSION_MAJOR < 11 - #error LLVM Version 11+ is required => "brew install llvm@11" - #endif - #if (LLVM_VERSION_MAJOR > 14 && LLVM_VERSION_MAJOR < 17) || LLVM_VERSION_MAJOR > 17 - #error LLVM Version 11..=14 or =17 is required => "brew install llvm@14" + #if LLVM_VERSION_MAJOR < 11 || (LLVM_VERSION_MAJOR > 14 && LLVM_VERSION_MAJOR < 17) || LLVM_VERSION_MAJOR > 18 + #error LLVM Version 11..=14 or =18 is required => "brew install llvm@14" #endif #endif @@ -272,6 +269,7 @@ enum BuildFlagKind { BuildFlag_ExtraAssemblerFlags, BuildFlag_Microarch, BuildFlag_TargetFeatures, + BuildFlag_StrictTargetFeatures, BuildFlag_MinimumOSVersion, BuildFlag_NoThreadLocal, @@ -467,6 +465,7 @@ gb_internal bool parse_build_flags(Array<String> args) { add_flag(&build_flags, BuildFlag_ExtraAssemblerFlags, str_lit("extra-assembler-flags"), BuildFlagParam_String, Command__does_build); add_flag(&build_flags, BuildFlag_Microarch, str_lit("microarch"), BuildFlagParam_String, Command__does_build); add_flag(&build_flags, BuildFlag_TargetFeatures, str_lit("target-features"), BuildFlagParam_String, Command__does_build); + add_flag(&build_flags, BuildFlag_StrictTargetFeatures, str_lit("strict-target-features"), BuildFlagParam_None, Command__does_build); add_flag(&build_flags, BuildFlag_MinimumOSVersion, str_lit("minimum-os-version"), BuildFlagParam_String, Command__does_build); add_flag(&build_flags, BuildFlag_RelocMode, str_lit("reloc-mode"), BuildFlagParam_String, Command__does_build); @@ -1083,6 +1082,9 @@ gb_internal bool parse_build_flags(Array<String> args) { string_to_lower(&build_context.target_features_string); break; } + case BuildFlag_StrictTargetFeatures: + build_context.strict_target_features = true; + break; case BuildFlag_MinimumOSVersion: { GB_ASSERT(value.kind == ExactValue_String); build_context.minimum_os_version_string = value.value_string; @@ -1981,7 +1983,20 @@ gb_internal void print_show_help(String const arg0, String const &command) { print_usage_line(2, "Examples:"); print_usage_line(3, "-microarch:sandybridge"); print_usage_line(3, "-microarch:native"); - print_usage_line(3, "-microarch:? for a list"); + print_usage_line(3, "-microarch:\"?\" for a list"); + print_usage_line(0, ""); + + print_usage_line(1, "-target-features:<string>"); + print_usage_line(2, "Specifies CPU features to enable on top of the enabled features implied by -microarch."); + print_usage_line(2, "Examples:"); + print_usage_line(3, "-target-features:atomics"); + print_usage_line(3, "-target-features:\"sse2,aes\""); + print_usage_line(3, "-target-features:\"?\" for a list"); + print_usage_line(0, ""); + + print_usage_line(1, "-strict-target-features"); + print_usage_line(2, "Makes @(enable_target_features=\"...\") behave the same way as @(require_target_features=\"...\")."); + print_usage_line(2, "This enforces that all generated code uses features supported by the combination of -target, -microarch, and -target-features."); print_usage_line(0, ""); print_usage_line(1, "-reloc-mode:<string>"); @@ -2663,7 +2678,7 @@ int main(int arg_count, char const **arg_ptr) { // Check chosen microarchitecture. If not found or ?, print list. bool print_microarch_list = true; - if (build_context.microarch.len == 0) { + if (build_context.microarch.len == 0 || build_context.microarch == str_lit("native")) { // Autodetect, no need to print list. print_microarch_list = false; } else { @@ -2680,6 +2695,11 @@ int main(int arg_count, char const **arg_ptr) { } } + // Set and check build paths... + if (!init_build_paths(init_filename)) { + return 1; + } + String default_march = get_default_microarchitecture(); if (print_microarch_list) { if (build_context.microarch != "?") { @@ -2703,13 +2723,57 @@ int main(int arg_count, char const **arg_ptr) { return 0; } - // Set and check build paths... - if (!init_build_paths(init_filename)) { - return 1; + String march = get_final_microarchitecture(); + String default_features = get_default_features(); + { + String_Iterator it = {default_features, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + string_set_add(&build_context.target_features_set, str); + } + } + + if (build_context.target_features_string.len != 0) { + String_Iterator target_it = {build_context.target_features_string, 0}; + for (;;) { + String item = string_split_iterator(&target_it, ','); + if (item == "") break; + + String invalid; + if (!check_target_feature_is_valid_for_target_arch(item, &invalid) && item != str_lit("help")) { + if (item != str_lit("?")) { + gb_printf_err("Unkown target feature '%.*s'.\n", LIT(invalid)); + } + gb_printf("Possible -target-features for target %.*s are:\n", LIT(target_arch_names[build_context.metrics.arch])); + gb_printf("\n"); + + String feature_list = target_features_list[build_context.metrics.arch]; + String_Iterator it = {feature_list, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (check_single_target_feature_is_valid(default_features, str)) { + if (has_ansi_terminal_colours()) { + gb_printf("\t%.*s\x1b[38;5;244m (implied by target microarch %.*s)\x1b[0m\n", LIT(str), LIT(march)); + } else { + gb_printf("\t%.*s (implied by current microarch %.*s)\n", LIT(str), LIT(march)); + } + } else { + gb_printf("\t%.*s\n", LIT(str)); + } + } + + return 1; + } + + string_set_add(&build_context.target_features_set, item); + } } if (build_context.show_debug_messages) { - debugf("Selected microarch: %.*s\n", LIT(default_march)); + debugf("Selected microarch: %.*s\n", LIT(march)); + debugf("Default microarch features: %.*s\n", LIT(default_features)); for_array(i, build_context.build_paths) { String build_path = path_to_string(heap_allocator(), build_context.build_paths[i]); debugf("build_paths[%ld]: %.*s\n", i, LIT(build_path)); diff --git a/src/parser.cpp b/src/parser.cpp index f4d3dc48d..6e859fe32 100644 --- a/src/parser.cpp +++ b/src/parser.cpp @@ -1,7 +1,7 @@ #include "parser_pos.cpp" gb_internal u64 ast_file_vet_flags(AstFile *f) { - if (f->vet_flags_set) { + if (f != nullptr && f->vet_flags_set) { return f->vet_flags; } return build_context.vet_flags; @@ -3499,6 +3499,10 @@ gb_internal Ast *parse_type(AstFile *f) { Token token = advance_token(f); syntax_error(token, "Expected a type"); return ast_bad_expr(f, token, f->curr_token); + } else if (type->kind == Ast_ParenExpr && + unparen_expr(type) == nullptr) { + syntax_error(type, "Expected a type within the parentheses"); + return ast_bad_expr(f, type->ParenExpr.open, type->ParenExpr.close); } return type; } @@ -5710,7 +5714,7 @@ gb_internal bool determine_path_from_string(BlockingMutex *file_mutex, Ast *node // working directory of the exe to the library search paths. // Static libraries can be linked directly with the full pathname // - if (node->kind == Ast_ForeignImportDecl && string_ends_with(file_str, str_lit(".so"))) { + if (node->kind == Ast_ForeignImportDecl && (string_ends_with(file_str, str_lit(".so")) || string_contains_string(file_str, str_lit(".so.")))) { *path = file_str; return true; } diff --git a/src/string.cpp b/src/string.cpp index b92dd589e..86eddeddd 100644 --- a/src/string.cpp +++ b/src/string.cpp @@ -328,6 +328,25 @@ gb_internal bool string_contains_char(String const &s, u8 c) { return false; } +gb_internal bool string_contains_string(String const &haystack, String const &needle) { + if (needle.len == 0) return true; + if (needle.len > haystack.len) return false; + + for (isize i = 0; i <= haystack.len - needle.len; i++) { + bool found = true; + for (isize j = 0; j < needle.len; j++) { + if (haystack[i + j] != needle[j]) { + found = false; + break; + } + } + if (found) { + return true; + } + } + return false; +} + gb_internal String filename_from_path(String s) { isize i = string_extension_position(s); if (i >= 0) { diff --git a/src/string_map.cpp b/src/string_map.cpp index 4de88bbf9..802bf5853 100644 --- a/src/string_map.cpp +++ b/src/string_map.cpp @@ -24,6 +24,296 @@ gb_internal gb_inline StringHashKey string_hash_string(String const &s) { return hash_key; } + +#if 1 /* old string map */ + +template <typename T> +struct StringMapEntry { + String key; + u32 hash; + MapIndex next; + T value; +}; + +template <typename T> +struct StringMap { + MapIndex * hashes; + usize hashes_count; + StringMapEntry<T> *entries; + u32 count; + u32 entries_capacity; +}; + + +template <typename T> gb_internal void string_map_init (StringMap<T> *h, usize capacity = 16); +template <typename T> gb_internal void string_map_destroy (StringMap<T> *h); + +template <typename T> gb_internal T * string_map_get (StringMap<T> *h, char const *key); +template <typename T> gb_internal T * string_map_get (StringMap<T> *h, String const &key); +template <typename T> gb_internal T * string_map_get (StringMap<T> *h, StringHashKey const &key); + +template <typename T> gb_internal T & string_map_must_get(StringMap<T> *h, char const *key); +template <typename T> gb_internal T & string_map_must_get(StringMap<T> *h, String const &key); +template <typename T> gb_internal T & string_map_must_get(StringMap<T> *h, StringHashKey const &key); + +template <typename T> gb_internal void string_map_set (StringMap<T> *h, char const *key, T const &value); +template <typename T> gb_internal void string_map_set (StringMap<T> *h, String const &key, T const &value); +template <typename T> gb_internal void string_map_set (StringMap<T> *h, StringHashKey const &key, T const &value); + +// template <typename T> gb_internal void string_map_remove (StringMap<T> *h, StringHashKey const &key); +template <typename T> gb_internal void string_map_clear (StringMap<T> *h); +template <typename T> gb_internal void string_map_grow (StringMap<T> *h); +template <typename T> gb_internal void string_map_reserve (StringMap<T> *h, usize new_count); + +gb_internal gbAllocator string_map_allocator(void) { + return heap_allocator(); +} + +template <typename T> +gb_internal gb_inline void string_map_init(StringMap<T> *h, usize capacity) { + capacity = next_pow2_isize(capacity); + string_map_reserve(h, capacity); +} + +template <typename T> +gb_internal gb_inline void string_map_destroy(StringMap<T> *h) { + gb_free(string_map_allocator(), h->hashes); + gb_free(string_map_allocator(), h->entries); +} + + +template <typename T> +gb_internal void string_map__resize_hashes(StringMap<T> *h, usize count) { + h->hashes_count = cast(u32)resize_array_raw(&h->hashes, string_map_allocator(), h->hashes_count, count, MAP_CACHE_LINE_SIZE); +} + + +template <typename T> +gb_internal void string_map__reserve_entries(StringMap<T> *h, usize capacity) { + h->entries_capacity = cast(u32)resize_array_raw(&h->entries, string_map_allocator(), h->entries_capacity, capacity, MAP_CACHE_LINE_SIZE); +} + + +template <typename T> +gb_internal MapIndex string_map__add_entry(StringMap<T> *h, u32 hash, String const &key) { + StringMapEntry<T> e = {}; + e.key = key; + e.hash = hash; + e.next = MAP_SENTINEL; + if (h->count+1 >= h->entries_capacity) { + string_map__reserve_entries(h, gb_max(h->entries_capacity*2, 4)); + } + h->entries[h->count++] = e; + return cast(MapIndex)(h->count-1); +} + +template <typename T> +gb_internal MapFindResult string_map__find(StringMap<T> *h, u32 hash, String const &key) { + MapFindResult fr = {MAP_SENTINEL, MAP_SENTINEL, MAP_SENTINEL}; + if (h->hashes_count != 0) { + fr.hash_index = cast(MapIndex)(hash & (h->hashes_count-1)); + fr.entry_index = h->hashes[fr.hash_index]; + while (fr.entry_index != MAP_SENTINEL) { + auto *entry = &h->entries[fr.entry_index]; + if (entry->hash == hash && entry->key == key) { + return fr; + } + fr.entry_prev = fr.entry_index; + fr.entry_index = entry->next; + } + } + return fr; +} + +template <typename T> +gb_internal MapFindResult string_map__find_from_entry(StringMap<T> *h, StringMapEntry<T> *e) { + MapFindResult fr = {MAP_SENTINEL, MAP_SENTINEL, MAP_SENTINEL}; + if (h->hashes_count != 0) { + fr.hash_index = cast(MapIndex)(e->hash & (h->hashes_count-1)); + fr.entry_index = h->hashes[fr.hash_index]; + while (fr.entry_index != MAP_SENTINEL) { + auto *entry = &h->entries[fr.entry_index]; + if (entry == e) { + return fr; + } + fr.entry_prev = fr.entry_index; + fr.entry_index = entry->next; + } + } + return fr; +} + +template <typename T> +gb_internal b32 string_map__full(StringMap<T> *h) { + return 0.75f * h->hashes_count <= h->count; +} + +template <typename T> +gb_inline void string_map_grow(StringMap<T> *h) { + isize new_count = gb_max(h->hashes_count<<1, 16); + string_map_reserve(h, new_count); +} + + +template <typename T> +gb_internal void string_map_reset_entries(StringMap<T> *h) { + for (u32 i = 0; i < h->hashes_count; i++) { + h->hashes[i] = MAP_SENTINEL; + } + for (isize i = 0; i < h->count; i++) { + MapFindResult fr; + StringMapEntry<T> *e = &h->entries[i]; + e->next = MAP_SENTINEL; + fr = string_map__find_from_entry(h, e); + if (fr.entry_prev == MAP_SENTINEL) { + h->hashes[fr.hash_index] = cast(MapIndex)i; + } else { + h->entries[fr.entry_prev].next = cast(MapIndex)i; + } + } +} + +template <typename T> +gb_internal void string_map_reserve(StringMap<T> *h, usize cap) { + if (h->count*2 < h->hashes_count) { + return; + } + string_map__reserve_entries(h, cap); + string_map__resize_hashes(h, cap*2); + string_map_reset_entries(h); +} + +template <typename T> +gb_internal T *string_map_get(StringMap<T> *h, u32 hash, String const &key) { + MapFindResult fr = {MAP_SENTINEL, MAP_SENTINEL, MAP_SENTINEL}; + if (h->hashes_count != 0) { + fr.hash_index = cast(MapIndex)(hash & (h->hashes_count-1)); + fr.entry_index = h->hashes[fr.hash_index]; + while (fr.entry_index != MAP_SENTINEL) { + auto *entry = &h->entries[fr.entry_index]; + if (entry->hash == hash && entry->key == key) { + return &entry->value; + } + fr.entry_prev = fr.entry_index; + fr.entry_index = entry->next; + } + } + return nullptr; +} + + +template <typename T> +gb_internal gb_inline T *string_map_get(StringMap<T> *h, StringHashKey const &key) { + return string_map_get(h, key.hash, key.string); +} + +template <typename T> +gb_internal gb_inline T *string_map_get(StringMap<T> *h, String const &key) { + return string_map_get(h, string_hash(key), key); +} + +template <typename T> +gb_internal gb_inline T *string_map_get(StringMap<T> *h, char const *key) { + String k = make_string_c(key); + return string_map_get(h, string_hash(k), k); +} + +template <typename T> +gb_internal T &string_map_must_get(StringMap<T> *h, u32 hash, String const &key) { + isize index = string_map__find(h, hash, key).entry_index; + GB_ASSERT(index != MAP_SENTINEL); + return h->entries[index].value; +} + +template <typename T> +gb_internal T &string_map_must_get(StringMap<T> *h, StringHashKey const &key) { + return string_map_must_get(h, key.hash, key.string); +} + +template <typename T> +gb_internal gb_inline T &string_map_must_get(StringMap<T> *h, String const &key) { + return string_map_must_get(h, string_hash(key), key); +} + +template <typename T> +gb_internal gb_inline T &string_map_must_get(StringMap<T> *h, char const *key) { + String k = make_string_c(key); + return string_map_must_get(h, string_hash(k), k); +} + +template <typename T> +gb_internal void string_map_set(StringMap<T> *h, u32 hash, String const &key, T const &value) { + MapIndex index; + MapFindResult fr; + if (h->hashes_count == 0) { + string_map_grow(h); + } + fr = string_map__find(h, hash, key); + if (fr.entry_index != MAP_SENTINEL) { + index = fr.entry_index; + } else { + index = string_map__add_entry(h, hash, key); + if (fr.entry_prev != MAP_SENTINEL) { + h->entries[fr.entry_prev].next = index; + } else { + h->hashes[fr.hash_index] = index; + } + } + h->entries[index].value = value; + + if (string_map__full(h)) { + string_map_grow(h); + } +} + +template <typename T> +gb_internal gb_inline void string_map_set(StringMap<T> *h, String const &key, T const &value) { + string_map_set(h, string_hash_string(key), value); +} + +template <typename T> +gb_internal gb_inline void string_map_set(StringMap<T> *h, char const *key, T const &value) { + string_map_set(h, string_hash_string(make_string_c(key)), value); +} + +template <typename T> +gb_internal gb_inline void string_map_set(StringMap<T> *h, StringHashKey const &key, T const &value) { + string_map_set(h, key.hash, key.string, value); +} + + +template <typename T> +gb_internal gb_inline void string_map_clear(StringMap<T> *h) { + h->count = 0; + for (u32 i = 0; i < h->hashes_count; i++) { + h->hashes[i] = MAP_SENTINEL; + } +} + + + +template <typename T> +gb_internal StringMapEntry<T> *begin(StringMap<T> &m) noexcept { + return m.entries; +} +template <typename T> +gb_internal StringMapEntry<T> const *begin(StringMap<T> const &m) noexcept { + return m.entries; +} + + +template <typename T> +gb_internal StringMapEntry<T> *end(StringMap<T> &m) noexcept { + return m.entries + m.count; +} + +template <typename T> +gb_internal StringMapEntry<T> const *end(StringMap<T> const &m) noexcept { + return m.entries + m.count; +} + +#else /* new string map */ + template <typename T> struct StringMapEntry { String key; @@ -305,3 +595,5 @@ gb_internal StringMapIterator<T> const begin(StringMap<T> const &m) noexcept { } return StringMapIterator<T>{&m, index}; } + +#endif
\ No newline at end of file diff --git a/src/threading.cpp b/src/threading.cpp index b2cfa6d8e..48c58e8f4 100644 --- a/src/threading.cpp +++ b/src/threading.cpp @@ -492,6 +492,8 @@ gb_internal u32 thread_current_id(void) { thread_id = gettid(); #elif defined(GB_SYSTEM_HAIKU) thread_id = find_thread(NULL); +#elif defined(GB_SYSTEM_FREEBSD) + thread_id = pthread_getthreadid_np(); #else #error Unsupported architecture for thread_current_id() #endif diff --git a/src/types.cpp b/src/types.cpp index 18cb12ea1..30e009086 100644 --- a/src/types.cpp +++ b/src/types.cpp @@ -184,6 +184,8 @@ struct TypeProc { isize specialization_count; ProcCallingConvention calling_convention; i32 variadic_index; + String require_target_feature; + String enable_target_feature; // TODO(bill): Make this a flag set rather than bools bool variadic; bool require_results; @@ -2991,7 +2993,22 @@ gb_internal Type *union_tag_type(Type *u) { return t_uint; } +gb_internal int matched_target_features(TypeProc *t) { + if (t->require_target_feature.len == 0) { + return 0; + } + int matches = 0; + String_Iterator it = {t->require_target_feature, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (check_target_feature_is_valid_for_target_arch(str, nullptr)) { + matches += 1; + } + } + return matches; +} enum ProcTypeOverloadKind { ProcOverload_Identical, // The types are identical @@ -3003,6 +3020,7 @@ enum ProcTypeOverloadKind { ProcOverload_ResultCount, ProcOverload_ResultTypes, ProcOverload_Polymorphic, + ProcOverload_TargetFeatures, ProcOverload_NotProcedure, @@ -3060,6 +3078,10 @@ gb_internal ProcTypeOverloadKind are_proc_types_overload_safe(Type *x, Type *y) } } + if (matched_target_features(&px) != matched_target_features(&py)) { + return ProcOverload_TargetFeatures; + } + if (px.params != nullptr && py.params != nullptr) { Entity *ex = px.params->Tuple.variables[0]; Entity *ey = py.params->Tuple.variables[0]; @@ -3245,6 +3267,10 @@ gb_internal Selection lookup_field_with_selection(Type *type_, String field_name } } + if (is_type_polymorphic(type)) { + // NOTE(bill): A polymorphic struct has no fields, this only hits in the case of an error + return sel; + } wait_signal_until_available(&type->Struct.fields_wait_signal); isize field_count = type->Struct.fields.count; if (field_count != 0) for_array(i, type->Struct.fields) { |