mirrors/zig
1
0
Fork 0
mirror of https://codeberg.org/ziglang/zig.git synced 2025-12-06 13:54:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

tools/update_cpu_features: add a "flatten" feature

Browse source 
and use it to clean up aarch64 target CPU features
This commit is contained in:
Andrew Kelley 2021-02-27 17:26:06 -07:00
parent e02acc0288
commit 85ab1b01e7
2 changed files with 455 additions and 163 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

388
lib/std/target/aarch64.zig
View file

@ -1,14 +1,10 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2015-2021 Zig Contributors
// This file is part of [zig](https://ziglang.org/), which is MIT licensed.
// The MIT license requires this copyright notice to be included in all copies
// and substantial portions of the software.
//! This file is auto-generated by tools/update_cpu_features.zig.
const std = @import("../std.zig");
const CpuFeature = std.Target.Cpu.Feature;
const CpuModel = std.Target.Cpu.Model;
pub const Feature = enum {
a34,
a65,
a76,
aes,
@ -17,8 +13,6 @@ pub const Feature = enum {
altnzcv,
am,
amvs,
apple_a10,
apple_a11,
apple_a12,
apple_a13,
apple_a7,
@ -26,6 +20,7 @@ pub const Feature = enum {
arith_cbz_fusion,
balance_fp_ops,
bf16,
brbe,
bti,
call_saved_x10,
call_saved_x11,
@ -39,8 +34,11 @@ pub const Feature = enum {
ccdp,
ccidx,
ccpp,
cmp_bcc_fusion,
complxnum,
context_id_rel2,
contextidr_el2,
cortex_a78c,
cortex_r82,
crc,
crypto,
custom_cheap_as_move,
@ -50,14 +48,14 @@ pub const Feature = enum {
ecv,
ete,
exynos_cheap_as_move,
exynosm4,
exynos_m4,
f32mm,
f64mm,
fgt,
fmi,
flagm,
force_32bit_jump_tables,
fp_armv8,
fp16fml,
fp_armv8,
fptoint,
fullfp16,
fuse_address,
@ -68,9 +66,11 @@ pub const Feature = enum {
fuse_literals,
harden_sls_blr,
harden_sls_retbr,
hcx,
i8mm,
jsconv,
lor,
ls64,
lse,
lsl_fast,
mpam,
@ -83,9 +83,9 @@ pub const Feature = enum {
no_neg_immediates,
nv,
outline_atomics,
pa,
pan,
pan_rwv,
pauth,
perfmon,
pmu,
predictable_select_expensive,
@ -129,6 +129,7 @@ pub const Feature = enum {
slow_strqro_store,
sm4,
spe,
spe_eef,
specrestrict,
ssbs,
strict_align,
@ -151,15 +152,18 @@ pub const Feature = enum {
use_experimental_zeroing_pseudos,
use_postra_scheduler,
use_reciprocal_square_root,
v8a,
v8_1a,
v8_2a,
v8_3a,
v8_4a,
v8_5a,
v8_6a,
v8_7a,
v8a,
v8r,
vh,
wfxt,
xs,
zcm,
zcz,
zcz_fp,
@ -174,26 +178,16 @@ pub const all_features = blk: {
const len = @typeInfo(Feature).Enum.fields.len;
std.debug.assert(len <= CpuFeature.Set.needed_bit_count);
var result: [len]CpuFeature = undefined;
result[@enumToInt(Feature.a34)] = .{
.llvm_name = "a35",
.description = "Cortex-A34 ARM processors",
.dependencies = featureSet(&[_]Feature{
.crc,
.crypto,
.perfmon,
.v8a,
}),
};
result[@enumToInt(Feature.a65)] = .{
.llvm_name = "a65",
.description = "Cortex-A65 ARM processors",
.dependencies = featureSet(&[_]Feature{
.crypto,
.dotprod,
.fp_armv8,
.fullfp16,
.neon,
.ras,
.fuse_address,
.fuse_aes,
.fuse_literals,
.rcpc,
.ssbs,
.v8_2a,
@ -206,6 +200,7 @@ pub const all_features = blk: {
.crypto,
.dotprod,
.fullfp16,
.fuse_aes,
.rcpc,
.ssbs,
.v8_2a,
@ -245,49 +240,6 @@ pub const all_features = blk: {
.am,
}),
};
result[@enumToInt(Feature.apple_a10)] = .{
.llvm_name = "apple-a10",
.description = "Apple A10",
.dependencies = featureSet(&[_]Feature{
.alternate_sextload_cvt_f32_pattern,
.arith_bcc_fusion,
.arith_cbz_fusion,
.crc,
.crypto,
.disable_latency_sched_heuristic,
.fp_armv8,
.fuse_aes,
.fuse_crypto_eor,
.lor,
.neon,
.pan,
.perfmon,
.rdm,
.vh,
.zcm,
.zcz,
}),
};
result[@enumToInt(Feature.apple_a11)] = .{
.llvm_name = "apple-a11",
.description = "Apple A11",
.dependencies = featureSet(&[_]Feature{
.alternate_sextload_cvt_f32_pattern,
.arith_bcc_fusion,
.arith_cbz_fusion,
.crypto,
.disable_latency_sched_heuristic,
.fp_armv8,
.fullfp16,
.fuse_aes,
.fuse_crypto_eor,
.neon,
.perfmon,
.v8_2a,
.zcm,
.zcz,
}),
};
result[@enumToInt(Feature.apple_a12)] = .{
.llvm_name = "apple-a12",
.description = "Apple A12",
@ -297,11 +249,9 @@ pub const all_features = blk: {
.arith_cbz_fusion,
.crypto,
.disable_latency_sched_heuristic,
.fp_armv8,
.fullfp16,
.fuse_aes,
.fuse_crypto_eor,
.neon,
.perfmon,
.v8_3a,
.zcm,
@ -317,12 +267,9 @@ pub const all_features = blk: {
.arith_cbz_fusion,
.crypto,
.disable_latency_sched_heuristic,
.fp_armv8,
.fp16fml,
.fullfp16,
.fuse_aes,
.fuse_crypto_eor,
.neon,
.perfmon,
.sha3,
.v8_4a,
@ -339,10 +286,8 @@ pub const all_features = blk: {
.arith_cbz_fusion,
.crypto,
.disable_latency_sched_heuristic,
.fp_armv8,
.fuse_aes,
.fuse_crypto_eor,
.neon,
.perfmon,
.zcm,
.zcz,
@ -369,6 +314,11 @@ pub const all_features = blk: {
.description = "Enable BFloat16 Extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.brbe)] = .{
.llvm_name = "brbe",
.description = "Enable Branch Record Buffer Extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.bti)] = .{
.llvm_name = "bti",
.description = "Enable Branch Target Identification",
@ -434,6 +384,11 @@ pub const all_features = blk: {
.description = "Enable v8.2 data Cache Clean to Point of Persistence",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.cmp_bcc_fusion)] = .{
.llvm_name = "cmp-bcc-fusion",
.description = "CPU fuses cmp+bcc operations",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.complxnum)] = .{
.llvm_name = "complxnum",
.description = "Enable v8.3-A Floating-point complex number support",
@ -441,11 +396,38 @@ pub const all_features = blk: {
.neon,
}),
};
result[@enumToInt(Feature.context_id_rel2)] = .{
result[@enumToInt(Feature.contextidr_el2)] = .{
.llvm_name = "CONTEXTIDREL2",
.description = "Enable RW operand CONTEXTIDR_EL2",
.description = "Enable RW operand Context ID Register (EL2)",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.cortex_a78c)] = .{
.llvm_name = "cortex-a78c",
.description = "Cortex-A78C ARM processors",
.dependencies = featureSet(&[_]Feature{
.cmp_bcc_fusion,
.crypto,
.dotprod,
.flagm,
.fp16fml,
.fuse_aes,
.pauth,
.perfmon,
.rcpc,
.spe,
.ssbs,
.use_postra_scheduler,
.v8_2a,
}),
};
result[@enumToInt(Feature.cortex_r82)] = .{
.llvm_name = "cortex-r82",
.description = "Cortex-R82 ARM Processors",
.dependencies = featureSet(&[_]Feature{
.use_postra_scheduler,
.v8r,
}),
};
result[@enumToInt(Feature.crc)] = .{
.llvm_name = "crc",
.description = "Enable ARMv8 CRC-32 checksum instructions",
@ -456,7 +438,6 @@ pub const all_features = blk: {
.description = "Enable cryptographic instructions",
.dependencies = featureSet(&[_]Feature{
.aes,
.neon,
.sha2,
}),
};
@ -499,7 +480,7 @@ pub const all_features = blk: {
.custom_cheap_as_move,
}),
};
result[@enumToInt(Feature.exynosm4)] = .{
result[@enumToInt(Feature.exynos_m4)] = .{
.llvm_name = "exynosm4",
.description = "Samsung Exynos-M4 processors",
.dependencies = featureSet(&[_]Feature{
@ -541,8 +522,8 @@ pub const all_features = blk: {
.description = "Enable fine grained virtualization traps extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.fmi)] = .{
.llvm_name = "fmi",
result[@enumToInt(Feature.flagm)] = .{
.llvm_name = "flagm",
.description = "Enable v8.4-A Flag Manipulation Instructions",
.dependencies = featureSet(&[_]Feature{}),
};
@ -551,11 +532,6 @@ pub const all_features = blk: {
.description = "Force jump table entries to be 32-bits wide except at MinSize",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.fp_armv8)] = .{
.llvm_name = "fp-armv8",
.description = "Enable ARMv8 FP",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.fp16fml)] = .{
.llvm_name = "fp16fml",
.description = "Enable FP16 FML instructions",
@ -563,6 +539,11 @@ pub const all_features = blk: {
.fullfp16,
}),
};
result[@enumToInt(Feature.fp_armv8)] = .{
.llvm_name = "fp-armv8",
.description = "Enable ARMv8 FP",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.fptoint)] = .{
.llvm_name = "fptoint",
.description = "Enable FRInt[32|64][Z|X] instructions that round a floating-point number to an integer (in FP format) forcing it to fit into a 32- or 64-bit int",
@ -615,6 +596,11 @@ pub const all_features = blk: {
.description = "Harden against straight line speculation across RET and BR instructions",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.hcx)] = .{
.llvm_name = "hcx",
.description = "Enable Armv8.7-A HCRX_EL2 system register",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.i8mm)] = .{
.llvm_name = "i8mm",
.description = "Enable Matrix Multiply Int8 Extension",
@ -632,6 +618,11 @@ pub const all_features = blk: {
.description = "Enables ARM v8.1 Limited Ordering Regions extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.ls64)] = .{
.llvm_name = "ls64",
.description = "Enable Armv8.7-A LD64B/ST64B Accelerator Extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.lse)] = .{
.llvm_name = "lse",
.description = "Enable ARMv8.1 Large System Extension (LSE) atomic instructions",
@ -665,9 +656,7 @@ pub const all_features = blk: {
.dependencies = featureSet(&[_]Feature{
.crypto,
.dotprod,
.fp_armv8,
.fullfp16,
.neon,
.rcpc,
.ssbs,
.v8_2a,
@ -679,9 +668,7 @@ pub const all_features = blk: {
.dependencies = featureSet(&[_]Feature{
.crypto,
.dotprod,
.fp_armv8,
.fullfp16,
.neon,
.rcpc,
.spe,
.ssbs,
@ -696,9 +683,7 @@ pub const all_features = blk: {
.ete,
.i8mm,
.mte,
.sve2,
.sve2_bitperm,
.trbe,
.v8_5a,
}),
};
@ -709,12 +694,9 @@ pub const all_features = blk: {
.bf16,
.ccdp,
.crypto,
.fp_armv8,
.fp16fml,
.fullfp16,
.fuse_aes,
.i8mm,
.neon,
.perfmon,
.rand,
.spe,
@ -731,7 +713,7 @@ pub const all_features = blk: {
};
result[@enumToInt(Feature.nv)] = .{
.llvm_name = "nv",
.description = "Enable v8.4-A Nested Virtualization extension",
.description = "Enable v8.4-A Nested Virtualization Enchancement",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.outline_atomics)] = .{
@ -739,11 +721,6 @@ pub const all_features = blk: {
.description = "Enable out of line atomics to support LSE instructions",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.pa)] = .{
.llvm_name = "pa",
.description = "Enable v8.3-A Pointer Authentication extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.pan)] = .{
.llvm_name = "pan",
.description = "Enables ARM v8.1 Privileged Access-Never extension",
@ -756,6 +733,11 @@ pub const all_features = blk: {
.pan,
}),
};
result[@enumToInt(Feature.pauth)] = .{
.llvm_name = "pauth",
.description = "Enable v8.3-A Pointer Authentication extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.perfmon)] = .{
.llvm_name = "perfmon",
.description = "Enable ARMv8 PMUv3 Performance Monitors extension",
@ -949,7 +931,6 @@ pub const all_features = blk: {
.llvm_name = "sha3",
.description = "Enable SHA512 and SHA3 support",
.dependencies = featureSet(&[_]Feature{
.neon,
.sha2,
}),
};
@ -980,6 +961,11 @@ pub const all_features = blk: {
.description = "Enable Statistical Profiling extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.spe_eef)] = .{
.llvm_name = "spe-eef",
.description = "Enable extra register in the Statistical Profiling Extension",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.specrestrict)] = .{
.llvm_name = "specrestrict",
.description = "Enable architectural speculation restriction",
@ -998,7 +984,9 @@ pub const all_features = blk: {
result[@enumToInt(Feature.sve)] = .{
.llvm_name = "sve",
.description = "Enable Scalable Vector Extension (SVE) instructions",
.dependencies = featureSet(&[_]Feature{}),
.dependencies = featureSet(&[_]Feature{
.fullfp16,
}),
};
result[@enumToInt(Feature.sve2)] = .{
.llvm_name = "sve2",
@ -1103,14 +1091,6 @@ pub const all_features = blk: {
.description = "Use the reciprocal square root approximation",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.v8a)] = .{
.llvm_name = null,
.description = "Support ARM v8a instructions",
.dependencies = featureSet(&[_]Feature{
.fp_armv8,
.neon,
}),
};
result[@enumToInt(Feature.v8_1a)] = .{
.llvm_name = "v8.1a",
.description = "Support ARM v8.1a instructions",
@ -1120,8 +1100,8 @@ pub const all_features = blk: {
.lse,
.pan,
.rdm,
.vh,
.v8a,
.vh,
}),
};
result[@enumToInt(Feature.v8_2a)] = .{
@ -1142,7 +1122,7 @@ pub const all_features = blk: {
.ccidx,
.complxnum,
.jsconv,
.pa,
.pauth,
.rcpc,
.v8_2a,
}),
@ -1154,7 +1134,7 @@ pub const all_features = blk: {
.am,
.dit,
.dotprod,
.fmi,
.flagm,
.mpam,
.nv,
.pmu,
@ -1192,13 +1172,73 @@ pub const all_features = blk: {
.v8_5a,
}),
};
result[@enumToInt(Feature.v8_7a)] = .{
.llvm_name = "v8.7a",
.description = "Support ARM v8.7a instructions",
.dependencies = featureSet(&[_]Feature{
.hcx,
.v8_6a,
.wfxt,
.xs,
}),
};
result[@enumToInt(Feature.v8a)] = .{
.llvm_name = null,
.description = "Support ARM v8a instructions",
.dependencies = featureSet(&[_]Feature{
.neon,
}),
};
result[@enumToInt(Feature.v8r)] = .{
.llvm_name = "v8r",
.description = "Support ARM v8r instructions",
.dependencies = featureSet(&[_]Feature{
.ccidx,
.ccpp,
.complxnum,
.contextidr_el2,
.crc,
.dit,
.dotprod,
.flagm,
.fp16fml,
.jsconv,
.lse,
.pan_rwv,
.pauth,
.perfmon,
.predres,
.ras,
.rcpc_immo,
.rdm,
.sb,
.sel2,
.sha3,
.sm4,
.specrestrict,
.ssbs,
.tlb_rmi,
.tracev8_4,
.uaops,
}),
};
result[@enumToInt(Feature.vh)] = .{
.llvm_name = "vh",
.description = "Enables ARM v8.1 Virtual Host extension",
.dependencies = featureSet(&[_]Feature{
.context_id_rel2,
.contextidr_el2,
}),
};
result[@enumToInt(Feature.wfxt)] = .{
.llvm_name = "wfxt",
.description = "Enable Armv8.7-A WFET and WFIT instruction",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.xs)] = .{
.llvm_name = "xs",
.description = "Enable Armv8.7-A limited-TLB-maintenance instruction",
.dependencies = featureSet(&[_]Feature{}),
};
result[@enumToInt(Feature.zcm)] = .{
.llvm_name = "zcm",
.description = "Has zero-cycle register moves",
@ -1240,11 +1280,11 @@ pub const cpu = struct {
.name = "a64fx",
.llvm_name = "a64fx",
.features = featureSet(&[_]Feature{
.aggressive_fma,
.arith_bcc_fusion,
.complxnum,
.fp_armv8,
.fullfp16,
.neon,
.perfmon,
.predictable_select_expensive,
.sha2,
.sve,
.use_postra_scheduler,
@ -1255,14 +1295,39 @@ pub const cpu = struct {
.name = "apple_a10",
.llvm_name = "apple-a10",
.features = featureSet(&[_]Feature{
.apple_a10,
.alternate_sextload_cvt_f32_pattern,
.arith_bcc_fusion,
.arith_cbz_fusion,
.crc,
.crypto,
.disable_latency_sched_heuristic,
.fuse_aes,
.fuse_crypto_eor,
.lor,
.pan,
.perfmon,
.rdm,
.vh,
.zcm,
.zcz,
}),
};
pub const apple_a11 = CpuModel{
.name = "apple_a11",
.llvm_name = "apple-a11",
.features = featureSet(&[_]Feature{
.apple_a11,
.alternate_sextload_cvt_f32_pattern,
.arith_bcc_fusion,
.arith_cbz_fusion,
.crypto,
.disable_latency_sched_heuristic,
.fullfp16,
.fuse_aes,
.fuse_crypto_eor,
.perfmon,
.v8_2a,
.zcm,
.zcz,
}),
};
pub const apple_a12 = CpuModel{
@ -1279,6 +1344,37 @@ pub const cpu = struct {
.apple_a13,
}),
};
pub const apple_a14 = CpuModel{
.name = "apple_a14",
.llvm_name = "apple-a14",
.features = featureSet(&[_]Feature{
.aggressive_fma,
.alternate_sextload_cvt_f32_pattern,
.altnzcv,
.arith_bcc_fusion,
.arith_cbz_fusion,
.ccdp,
.crypto,
.disable_latency_sched_heuristic,
.fp16fml,
.fptoint,
.fuse_address,
.fuse_aes,
.fuse_arith_logic,
.fuse_crypto_eor,
.fuse_csel,
.fuse_literals,
.perfmon,
.predres,
.sb,
.sha3,
.specrestrict,
.ssbs,
.v8_4a,
.zcm,
.zcz,
}),
};
pub const apple_a7 = CpuModel{
.name = "apple_a7",
.llvm_name = "apple-a7",
@ -1327,7 +1423,6 @@ pub const cpu = struct {
.features = featureSet(&[_]Feature{
.crypto,
.fullfp16,
.neon,
.v8_2a,
}),
};
@ -1335,14 +1430,20 @@ pub const cpu = struct {
.name = "cortex_a34",
.llvm_name = "cortex-a34",
.features = featureSet(&[_]Feature{
.a34,
.crc,
.crypto,
.perfmon,
.v8a,
}),
};
pub const cortex_a35 = CpuModel{
.name = "cortex_a35",
.llvm_name = "cortex-a35",
.features = featureSet(&[_]Feature{
.a34,
.crc,
.crypto,
.perfmon,
.v8a,
}),
};
pub const cortex_a53 = CpuModel{
@ -1410,6 +1511,7 @@ pub const cpu = struct {
.crc,
.crypto,
.fuse_aes,
.fuse_literals,
.perfmon,
.v8a,
}),
@ -1456,11 +1558,11 @@ pub const cpu = struct {
.name = "cortex_a77",
.llvm_name = "cortex-a77",
.features = featureSet(&[_]Feature{
.cmp_bcc_fusion,
.crypto,
.dotprod,
.fp_armv8,
.fullfp16,
.neon,
.fuse_aes,
.rcpc,
.v8_2a,
}),
@ -1469,12 +1571,11 @@ pub const cpu = struct {
.name = "cortex_a78",
.llvm_name = "cortex-a78",
.features = featureSet(&[_]Feature{
.cmp_bcc_fusion,
.crypto,
.dotprod,
.fp_armv8,
.fullfp16,
.fuse_aes,
.neon,
.perfmon,
.rcpc,
.spe,
@ -1483,24 +1584,29 @@ pub const cpu = struct {
.v8_2a,
}),
};
pub const cortex_a78c = CpuModel{
.name = "cortex_a78c",
.llvm_name = "cortex-a78c",
.features = featureSet(&[_]Feature{
.cortex_a78c,
}),
};
pub const cortex_r82 = CpuModel{
.name = "cortex_r82",
.llvm_name = "cortex-r82",
.features = featureSet(&[_]Feature{
.use_postra_scheduler,
.v8r,
.cortex_r82,
}),
};
pub const cortex_x1 = CpuModel{
.name = "cortex_x1",
.llvm_name = "cortex-x1",
.features = featureSet(&[_]Feature{
.cmp_bcc_fusion,
.crypto,
.dotprod,
.fp_armv8,
.fullfp16,
.fuse_aes,
.neon,
.perfmon,
.rcpc,
.spe,
@ -1574,14 +1680,14 @@ pub const cpu = struct {
.name = "exynos_m4",
.llvm_name = "exynos-m4",
.features = featureSet(&[_]Feature{
.exynosm4,
.exynos_m4,
}),
};
pub const exynos_m5 = CpuModel{
.name = "exynos_m5",
.llvm_name = "exynos-m5",
.features = featureSet(&[_]Feature{
.exynosm4,
.exynos_m4,
}),
};
pub const falkor = CpuModel{
@ -1607,9 +1713,9 @@ pub const cpu = struct {
.features = featureSet(&[_]Feature{
.ete,
.fuse_aes,
.neon,
.perfmon,
.use_postra_scheduler,
.v8a,
}),
};
pub const kryo = CpuModel{
@ -1623,8 +1729,8 @@ pub const cpu = struct {
.perfmon,
.predictable_select_expensive,
.use_postra_scheduler,
.zcz,
.v8a,
.zcz,
}),
};
pub const neoverse_e1 = CpuModel{
@ -1688,9 +1794,7 @@ pub const cpu = struct {
.features = featureSet(&[_]Feature{
.aggressive_fma,
.arith_bcc_fusion,
.crc,
.crypto,
.lse,
.predictable_select_expensive,
.use_postra_scheduler,
.v8_1a,
@ -1703,12 +1807,7 @@ pub const cpu = struct {
.aggressive_fma,
.arith_bcc_fusion,
.balance_fp_ops,
.crc,
.crypto,
.fp_armv8,
.lse,
.neon,
.pa,
.perfmon,
.predictable_select_expensive,
.strict_align,
@ -1761,7 +1860,6 @@ pub const cpu = struct {
.custom_cheap_as_move,
.dotprod,
.fp16fml,
.fullfp16,
.fuse_aes,
.perfmon,
.spe,

228
tools/update_cpu_features.zig
View file

@ -8,7 +8,11 @@ const assert = std.debug.assert;
const FeatureOverride = struct {
llvm_name: []const u8,
/// If true, completely omit the feature; as if it does not exist.
omit: bool = false,
/// If true, omit the feature, but all the dependencies of the feature
/// are added in its place.
flatten: bool = false,
zig_name: ?[]const u8 = null,
desc: ?[]const u8 = null,
extra_deps: []const []const u8 = &.{},
@ -25,6 +29,7 @@ const Feature = struct {
zig_name: []const u8,
desc: []const u8,
deps: []const []const u8,
flatten: bool = false,
};
const LlvmTarget = struct {
@ -68,6 +73,8 @@ const llvm_targets = [_]LlvmTarget{
.{
.llvm_name = "exynosm3",
.zig_name = "exynos_m3",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "exynosm4",
@ -77,6 +84,117 @@ const llvm_targets = [_]LlvmTarget{
.llvm_name = "v8.1a",
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "a35",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "a53",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "a55",
.flatten = true,
},
.{
.llvm_name = "a57",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "a64fx",
.flatten = true,
},
.{
.llvm_name = "a72",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "a73",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "a75",
.flatten = true,
},
.{
.llvm_name = "a77",
.flatten = true,
},
.{
.llvm_name = "apple-a10",
.flatten = true,
},
.{
.llvm_name = "apple-a11",
.flatten = true,
},
.{
.llvm_name = "apple-a14",
.flatten = true,
},
.{
.llvm_name = "carmel",
.flatten = true,
},
.{
.llvm_name = "cortex-a78",
.flatten = true,
},
.{
.llvm_name = "cortex-x1",
.flatten = true,
},
.{
.llvm_name = "falkor",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "kryo",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "saphira",
.flatten = true,
},
.{
.llvm_name = "thunderx",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "thunderx2t99",
.flatten = true,
},
.{
.llvm_name = "thunderx3t110",
.flatten = true,
},
.{
.llvm_name = "thunderxt81",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "thunderxt83",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "thunderxt88",
.flatten = true,
.extra_deps = &.{"v8a"},
},
.{
.llvm_name = "tsv110",
.flatten = true,
},
},
.extra_features = &.{
.{
@ -85,6 +203,43 @@ const llvm_targets = [_]LlvmTarget{
.deps = &.{ "fp_armv8", "neon" },
},
},
.extra_cpus = &.{
.{
.llvm_name = null,
.zig_name = "exynos_m1",
.features = &.{
"crc",
"crypto",
"exynos_cheap_as_move",
"force_32bit_jump_tables",
"fuse_aes",
"perfmon",
"slow_misaligned_128store",
"slow_paired_128",
"use_postra_scheduler",
"use_reciprocal_square_root",
"v8a",
"zcz_fp",
},
},
.{
.llvm_name = null,
.zig_name = "exynos_m2",
.features = &.{
"crc",
"crypto",
"exynos_cheap_as_move",
"force_32bit_jump_tables",
"fuse_aes",
"perfmon",
"slow_misaligned_128store",
"slow_paired_128",
"use_postra_scheduler",
"v8a",
"zcz_fp",
},
},
},
},
.{
.zig_name = "amdgpu",
@ -374,22 +529,32 @@ fn processOneTarget(job: Job) anyerror!void {
const llvm_name = kv.value.Object.get("Name").?.String;
if (llvm_name.len == 0) continue;
var zig_name = (try llvmNameToZigName(arena, llvm_target, llvm_name)) orelse
continue :root_it;
var zig_name = try llvmNameToZigName(arena, llvm_name);
var desc = kv.value.Object.get("Desc").?.String;
var deps = std.ArrayList([]const u8).init(arena);
var omit = false;
var flatten = false;
const implies = kv.value.Object.get("Implies").?.Array;
for (implies.items) |imply| {
const other_key = imply.Object.get("def").?.String;
const other_obj = &root_map.getEntry(other_key).?.value.Object;
const other_llvm_name = other_obj.get("Name").?.String;
const other_zig_name = (try llvmNameToZigName(arena, llvm_target, other_llvm_name)) orelse continue;
const other_zig_name = (try llvmNameToZigNameOmit(
arena,
llvm_target,
other_llvm_name,
)) orelse continue;
try deps.append(other_zig_name);
}
for (llvm_target.feature_overrides) |feature_override| {
if (mem.eql(u8, llvm_name, feature_override.llvm_name)) {
if (feature_override.omit) {
continue :root_it;
// Still put the feature into the table so that we can
// expand dependencies for the feature overrides marked `flatten`.
omit = true;
}
if (feature_override.flatten) {
flatten = true;
}
if (feature_override.zig_name) |override_name| {
zig_name = override_name;
@ -408,16 +573,18 @@ fn processOneTarget(job: Job) anyerror!void {
.zig_name = zig_name,
.desc = desc,
.deps = deps.items,
.flatten = flatten,
};
try features_table.put(zig_name, feature);
if (!omit and !flatten) {
try all_features.append(feature);
}
}
if (hasSuperclass(&kv.value.Object, "Processor")) {
const llvm_name = kv.value.Object.get("Name").?.String;
if (llvm_name.len == 0) continue;
var zig_name = (try llvmNameToZigName(arena, llvm_target, llvm_name)) orelse
continue :root_it;
var zig_name = try llvmNameToZigName(arena, llvm_name);
var deps = std.ArrayList([]const u8).init(arena);
const features = kv.value.Object.get("Features").?.Array;
for (features.items) |feature| {
@ -425,7 +592,11 @@ fn processOneTarget(job: Job) anyerror!void {
const feature_obj = &root_map.getEntry(feature_key).?.value.Object;
const feature_llvm_name = feature_obj.get("Name").?.String;
if (feature_llvm_name.len == 0) continue;
const feature_zig_name = (try llvmNameToZigName(arena, llvm_target, feature_llvm_name)) orelse continue;
const feature_zig_name = (try llvmNameToZigNameOmit(
arena,
llvm_target,
feature_llvm_name,
)) orelse continue;
try deps.append(feature_zig_name);
}
const tune_features = kv.value.Object.get("TuneFeatures").?.Array;
@ -434,7 +605,11 @@ fn processOneTarget(job: Job) anyerror!void {
const feature_obj = &root_map.getEntry(feature_key).?.value.Object;
const feature_llvm_name = feature_obj.get("Name").?.String;
if (feature_llvm_name.len == 0) continue;
const feature_zig_name = (try llvmNameToZigName(arena, llvm_target, feature_llvm_name)) orelse continue;
const feature_zig_name = (try llvmNameToZigNameOmit(
arena,
llvm_target,
feature_llvm_name,
)) orelse continue;
try deps.append(feature_zig_name);
}
for (llvm_target.feature_overrides) |feature_override| {
@ -549,7 +724,7 @@ fn processOneTarget(job: Job) anyerror!void {
}
var deps_set = std.StringHashMap(void).init(arena);
for (feature.deps) |dep| {
try deps_set.put(dep, {});
try putDep(&deps_set, features_table, dep);
}
try pruneFeatures(arena, features_table, &deps_set);
var dependencies = std.ArrayList([]const u8).init(arena);
@ -594,7 +769,7 @@ fn processOneTarget(job: Job) anyerror!void {
for (all_cpus.items) |cpu| {
var deps_set = std.StringHashMap(void).init(arena);
for (cpu.features) |feature_zig_name| {
try deps_set.put(feature_zig_name, {});
try putDep(&deps_set, features_table, feature_zig_name);
}
try pruneFeatures(arena, features_table, &deps_set);
var cpu_features = std.ArrayList([]const u8).init(arena);
@ -679,7 +854,16 @@ fn asciiLessThan(context: void, a: []const u8, b: []const u8) bool {
return std.ascii.lessThanIgnoreCase(a, b);
}
fn llvmNameToZigName(
fn llvmNameToZigName(arena: *mem.Allocator, llvm_name: []const u8) ![]const u8 {
const duped = try arena.dupe(u8, llvm_name);
for (duped) |*byte| switch (byte.*) {
'-', '.' => byte.* = '_',
else => continue,
};
return duped;
}
fn llvmNameToZigNameOmit(
arena: *mem.Allocator,
llvm_target: LlvmTarget,
llvm_name: []const u8,
@ -690,12 +874,7 @@ fn llvmNameToZigName(
return feature_override.zig_name orelse break;
}
}
const duped = try arena.dupe(u8, llvm_name);
for (duped) |*byte| switch (byte.*) {
'-', '.' => byte.* = '_',
else => continue,
};
return duped;
return try llvmNameToZigName(arena, llvm_name);
}
fn hasSuperclass(obj: *json.ObjectMap, class_name: []const u8) bool {
@ -744,3 +923,18 @@ fn walkFeatures(
try walkFeatures(features_table, deletion_set, other_feature);
}
}
fn putDep(
deps_set: *std.StringHashMap(void),
features_table: std.StringHashMap(Feature),
zig_feature_name: []const u8,
) error{OutOfMemory}!void {
const feature = features_table.get(zig_feature_name).?;
if (feature.flatten) {
for (feature.deps) |dep| {
try putDep(deps_set, features_table, dep);
}
} else {
try deps_set.put(zig_feature_name, {});
}
}