maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[X86] Remove avx512f integer and/or/xor/min/max reduction intrinsics and use generic equivalents 

None of these have any reordering issues, and they still emit the same reduction intrinsics without any change in the existing test coverage:

llvm-project\clang\test\CodeGen\X86\avx512-reduceIntrin.c
llvm-project\clang\test\CodeGen\X86\avx512-reduceMinMaxIntrin.c

Differential Revision: https://reviews.llvm.org/D117881
This commit is contained in:
Simon Pilgrim 2022-01-24 11:57:23 +00:00
parent 3696c70e67
commit e4074432d5
3 changed files with 24 additions and 72 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
clang/include/clang/Basic/BuiltinsX86.def
View File

@ -2015,8 +2015,6 @@ TARGET_BUILTIN(__builtin_ia32_selectsd_128, "V2dUcV2dV2d", "ncV:128:", "avx512f"
// generic reduction intrinsics
TARGET_BUILTIN(__builtin_ia32_reduce_add_d512, "iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_add_q512, "OiV8Oi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_and_d512, "iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_and_q512, "OiV8Oi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_fadd_pd512, "ddV8d", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_fadd_ps512, "ffV16f", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_fadd_ph512, "xxV32x", "ncV:512:", "avx512fp16")
@ -2039,16 +2037,6 @@ TARGET_BUILTIN(__builtin_ia32_reduce_fmul_ph256, "xxV16x", "ncV:256:", "avx512fp
TARGET_BUILTIN(__builtin_ia32_reduce_fmul_ph128, "xxV8x", "ncV:128:", "avx512fp16,avx512vl")
TARGET_BUILTIN(__builtin_ia32_reduce_mul_d512, "iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_mul_q512, "OiV8Oi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_or_d512, "iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_or_q512, "OiV8Oi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_smax_d512, "iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_smax_q512, "OiV8Oi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_smin_d512, "iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_smin_q512, "OiV8Oi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_umax_d512, "iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_umax_q512, "OiV8Oi", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_umin_d512, "iV16i", "ncV:512:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_reduce_umin_q512, "OiV8Oi", "ncV:512:", "avx512f")
// MONITORX/MWAITX
TARGET_BUILTIN(__builtin_ia32_monitorx, "vvC*UiUi", "n", "mwaitx")

36
clang/lib/CodeGen/CGBuiltin.cpp
View File

@ -14365,12 +14365,6 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
CGM.getIntrinsic(Intrinsic::vector_reduce_add, Ops[0]->getType());
return Builder.CreateCall(F, {Ops[0]});
}
case X86::BI__builtin_ia32_reduce_and_d512:
case X86::BI__builtin_ia32_reduce_and_q512: {
Function *F =
CGM.getIntrinsic(Intrinsic::vector_reduce_and, Ops[0]->getType());
return Builder.CreateCall(F, {Ops[0]});
}
case X86::BI__builtin_ia32_reduce_fadd_pd512:
case X86::BI__builtin_ia32_reduce_fadd_ps512:
case X86::BI__builtin_ia32_reduce_fadd_ph512:
@ -14417,36 +14411,6 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
CGM.getIntrinsic(Intrinsic::vector_reduce_mul, Ops[0]->getType());
return Builder.CreateCall(F, {Ops[0]});
}
case X86::BI__builtin_ia32_reduce_or_d512:
case X86::BI__builtin_ia32_reduce_or_q512: {
Function *F =
CGM.getIntrinsic(Intrinsic::vector_reduce_or, Ops[0]->getType());
return Builder.CreateCall(F, {Ops[0]});
}
case X86::BI__builtin_ia32_reduce_smax_d512:
case X86::BI__builtin_ia32_reduce_smax_q512: {
Function *F =
CGM.getIntrinsic(Intrinsic::vector_reduce_smax, Ops[0]->getType());
return Builder.CreateCall(F, {Ops[0]});
}
case X86::BI__builtin_ia32_reduce_smin_d512:
case X86::BI__builtin_ia32_reduce_smin_q512: {
Function *F =
CGM.getIntrinsic(Intrinsic::vector_reduce_smin, Ops[0]->getType());
return Builder.CreateCall(F, {Ops[0]});
}
case X86::BI__builtin_ia32_reduce_umax_d512:
case X86::BI__builtin_ia32_reduce_umax_q512: {
Function *F =
CGM.getIntrinsic(Intrinsic::vector_reduce_umax, Ops[0]->getType());
return Builder.CreateCall(F, {Ops[0]});
}
case X86::BI__builtin_ia32_reduce_umin_d512:
case X86::BI__builtin_ia32_reduce_umin_q512: {
Function *F =
CGM.getIntrinsic(Intrinsic::vector_reduce_umin, Ops[0]->getType());
return Builder.CreateCall(F, {Ops[0]});
}
// 3DNow!
case X86::BI__builtin_ia32_pswapdsf:

48
clang/lib/Headers/avx512fintrin.h
View File

@ -9324,11 +9324,11 @@ static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_mul_epi64(__m512
}
static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_and_epi64(__m512i __W) {
return __builtin_ia32_reduce_and_q512(__W);
return __builtin_reduce_and((__v8di)__W);
}
static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_or_epi64(__m512i __W) {
return __builtin_ia32_reduce_or_q512(__W);
return __builtin_reduce_or((__v8di)__W);
}
static __inline__ long long __DEFAULT_FN_ATTRS512
@ -9346,13 +9346,13 @@ _mm512_mask_reduce_mul_epi64(__mmask8 __M, __m512i __W) {
static __inline__ long long __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_and_epi64(__mmask8 __M, __m512i __W) {
__W = _mm512_mask_mov_epi64(_mm512_set1_epi64(~0ULL), __M, __W);
return __builtin_ia32_reduce_and_q512(__W);
return __builtin_reduce_and((__v8di)__W);
}
static __inline__ long long __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_or_epi64(__mmask8 __M, __m512i __W) {
__W = _mm512_maskz_mov_epi64(__M, __W);
return __builtin_ia32_reduce_or_q512(__W);
return __builtin_reduce_or((__v8di)__W);
}
// -0.0 is used to ignore the start value since it is the neutral value of
@ -9390,12 +9390,12 @@ _mm512_reduce_mul_epi32(__m512i __W) {
static __inline__ int __DEFAULT_FN_ATTRS512
_mm512_reduce_and_epi32(__m512i __W) {
return __builtin_ia32_reduce_and_d512((__v16si)__W);
return __builtin_reduce_and((__v16si)__W);
}
static __inline__ int __DEFAULT_FN_ATTRS512
_mm512_reduce_or_epi32(__m512i __W) {
return __builtin_ia32_reduce_or_d512((__v16si)__W);
return __builtin_reduce_or((__v16si)__W);
}
static __inline__ int __DEFAULT_FN_ATTRS512
@ -9413,13 +9413,13 @@ _mm512_mask_reduce_mul_epi32( __mmask16 __M, __m512i __W) {
static __inline__ int __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_and_epi32( __mmask16 __M, __m512i __W) {
__W = _mm512_mask_mov_epi32(_mm512_set1_epi32(~0U), __M, __W);
return __builtin_ia32_reduce_and_d512((__v16si)__W);
return __builtin_reduce_and((__v16si)__W);
}
static __inline__ int __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_or_epi32(__mmask16 __M, __m512i __W) {
__W = _mm512_maskz_mov_epi32(__M, __W);
return __builtin_ia32_reduce_or_d512((__v16si)__W);
return __builtin_reduce_or((__v16si)__W);
}
static __inline__ float __DEFAULT_FN_ATTRS512
@ -9446,89 +9446,89 @@ _mm512_mask_reduce_mul_ps(__mmask16 __M, __m512 __W) {
static __inline__ long long __DEFAULT_FN_ATTRS512
_mm512_reduce_max_epi64(__m512i __V) {
return __builtin_ia32_reduce_smax_q512(__V);
return __builtin_reduce_max((__v8di)__V);
}
static __inline__ unsigned long long __DEFAULT_FN_ATTRS512
_mm512_reduce_max_epu64(__m512i __V) {
return __builtin_ia32_reduce_umax_q512(__V);
return __builtin_reduce_max((__v8du)__V);
}
static __inline__ long long __DEFAULT_FN_ATTRS512
_mm512_reduce_min_epi64(__m512i __V) {
return __builtin_ia32_reduce_smin_q512(__V);
return __builtin_reduce_min((__v8di)__V);
}
static __inline__ unsigned long long __DEFAULT_FN_ATTRS512
_mm512_reduce_min_epu64(__m512i __V) {
return __builtin_ia32_reduce_umin_q512(__V);
return __builtin_reduce_min((__v8du)__V);
}
static __inline__ long long __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_max_epi64(__mmask8 __M, __m512i __V) {
__V = _mm512_mask_mov_epi64(_mm512_set1_epi64(-__LONG_LONG_MAX__ - 1LL), __M, __V);
return __builtin_ia32_reduce_smax_q512(__V);
return __builtin_reduce_max((__v8di)__V);
}
static __inline__ unsigned long long __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_max_epu64(__mmask8 __M, __m512i __V) {
__V = _mm512_maskz_mov_epi64(__M, __V);
return __builtin_ia32_reduce_umax_q512(__V);
return __builtin_reduce_max((__v8du)__V);
}
static __inline__ long long __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_min_epi64(__mmask8 __M, __m512i __V) {
__V = _mm512_mask_mov_epi64(_mm512_set1_epi64(__LONG_LONG_MAX__), __M, __V);
return __builtin_ia32_reduce_smin_q512(__V);
return __builtin_reduce_min((__v8di)__V);
}
static __inline__ unsigned long long __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_min_epu64(__mmask8 __M, __m512i __V) {
__V = _mm512_mask_mov_epi64(_mm512_set1_epi64(~0ULL), __M, __V);
return __builtin_ia32_reduce_umin_q512(__V);
return __builtin_reduce_min((__v8du)__V);
}
static __inline__ int __DEFAULT_FN_ATTRS512
_mm512_reduce_max_epi32(__m512i __V) {
return __builtin_ia32_reduce_smax_d512((__v16si)__V);
return __builtin_reduce_max((__v16si)__V);
}
static __inline__ unsigned int __DEFAULT_FN_ATTRS512
_mm512_reduce_max_epu32(__m512i __V) {
return __builtin_ia32_reduce_umax_d512((__v16si)__V);
return __builtin_reduce_max((__v16su)__V);
}
static __inline__ int __DEFAULT_FN_ATTRS512
_mm512_reduce_min_epi32(__m512i __V) {
return __builtin_ia32_reduce_smin_d512((__v16si)__V);
return __builtin_reduce_min((__v16si)__V);
}
static __inline__ unsigned int __DEFAULT_FN_ATTRS512
_mm512_reduce_min_epu32(__m512i __V) {
return __builtin_ia32_reduce_umin_d512((__v16si)__V);
return __builtin_reduce_min((__v16su)__V);
}
static __inline__ int __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_max_epi32(__mmask16 __M, __m512i __V) {
__V = _mm512_mask_mov_epi32(_mm512_set1_epi32(-__INT_MAX__ - 1), __M, __V);
return __builtin_ia32_reduce_smax_d512((__v16si)__V);
return __builtin_reduce_max((__v16si)__V);
}
static __inline__ unsigned int __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_max_epu32(__mmask16 __M, __m512i __V) {
__V = _mm512_maskz_mov_epi32(__M, __V);
return __builtin_ia32_reduce_umax_d512((__v16si)__V);
return __builtin_reduce_max((__v16su)__V);
}
static __inline__ int __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_min_epi32(__mmask16 __M, __m512i __V) {
__V = _mm512_mask_mov_epi32(_mm512_set1_epi32(__INT_MAX__), __M, __V);
return __builtin_ia32_reduce_smin_d512((__v16si)__V);
return __builtin_reduce_min((__v16si)__V);
}
static __inline__ unsigned int __DEFAULT_FN_ATTRS512
_mm512_mask_reduce_min_epu32(__mmask16 __M, __m512i __V) {
__V = _mm512_mask_mov_epi32(_mm512_set1_epi32(~0U), __M, __V);
return __builtin_ia32_reduce_umin_d512((__v16si)__V);
return __builtin_reduce_min((__v16su)__V);
}
static __inline__ double __DEFAULT_FN_ATTRS512