replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[X86] Remove the cvtuqq2ps256/cvtqq2ps256 mask builtins. Replace with uitofp/sitofp and select. 

Reviewers: RKSimon, spatel

Reviewed By: RKSimon

Subscribers: kristina, cfe-commits

Differential Revision: https://reviews.llvm.org/D56965

llvm-svn: 351694
This commit is contained in:
Craig Topper 2019-01-20 19:04:56 +00:00
parent ff6a9a275b
commit 9fddc3fd00
3 changed files with 24 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
clang/include/clang/Basic/BuiltinsX86.def
View File

@ -1301,7 +1301,6 @@ TARGET_BUILTIN(__builtin_ia32_cvtps2qq256_mask, "V4LLiV4fV4LLiUc", "ncV:256:", "
TARGET_BUILTIN(__builtin_ia32_cvtps2uqq128_mask, "V2LLiV4fV2LLiUc", "ncV:128:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvtps2uqq256_mask, "V4LLiV4fV4LLiUc", "ncV:256:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvtqq2ps128_mask, "V4fV2LLiV4fUc", "ncV:128:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvtqq2ps256_mask, "V4fV4LLiV4fUc", "ncV:256:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvttpd2qq128_mask, "V2LLiV2dV2LLiUc", "ncV:128:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvttpd2qq256_mask, "V4LLiV4dV4LLiUc", "ncV:256:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvttpd2uqq128_mask, "V2LLiV2dV2LLiUc", "ncV:128:", "avx512vl,avx512dq")
@ -1311,7 +1310,6 @@ TARGET_BUILTIN(__builtin_ia32_cvttps2qq256_mask, "V4LLiV4fV4LLiUc", "ncV:256:",
TARGET_BUILTIN(__builtin_ia32_cvttps2uqq128_mask, "V2LLiV4fV2LLiUc", "ncV:128:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvttps2uqq256_mask, "V4LLiV4fV4LLiUc", "ncV:256:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvtuqq2ps128_mask, "V4fV2LLiV4fUc", "ncV:128:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvtuqq2ps256_mask, "V4fV4LLiV4fUc", "ncV:256:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_rangepd128_mask, "V2dV2dV2dIiV2dUc", "ncV:128:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_rangepd256_mask, "V4dV4dV4dIiV4dUc", "ncV:256:", "avx512vl,avx512dq")
TARGET_BUILTIN(__builtin_ia32_rangeps128_mask, "V4fV4fV4fIiV4fUc", "ncV:128:", "avx512vl,avx512dq")

32
clang/lib/Headers/avx512vldqintrin.h
View File

@ -523,23 +523,21 @@ _mm_maskz_cvtepi64_ps (__mmask8 __U, __m128i __A) {
static __inline__ __m128 __DEFAULT_FN_ATTRS256
_mm256_cvtepi64_ps (__m256i __A) {
return (__m128) __builtin_ia32_cvtqq2ps256_mask ((__v4di) __A,
(__v4sf) _mm_setzero_ps(),
(__mmask8) -1);
return (__m128)__builtin_convertvector((__v4di)__A, __v4sf);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS256
_mm256_mask_cvtepi64_ps (__m128 __W, __mmask8 __U, __m256i __A) {
return (__m128) __builtin_ia32_cvtqq2ps256_mask ((__v4di) __A,
(__v4sf) __W,
(__mmask8) __U);
return (__m128)__builtin_ia32_selectps_128((__mmask8)__U,
(__v4sf)_mm256_cvtepi64_ps(__A),
(__v4sf)__W);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS256
_mm256_maskz_cvtepi64_ps (__mmask8 __U, __m256i __A) {
return (__m128) __builtin_ia32_cvtqq2ps256_mask ((__v4di) __A,
(__v4sf) _mm_setzero_ps(),
(__mmask8) __U);
return (__m128)__builtin_ia32_selectps_128((__mmask8)__U,
(__v4sf)_mm256_cvtepi64_ps(__A),
(__v4sf)_mm_setzero_ps());
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
@ -771,23 +769,21 @@ _mm_maskz_cvtepu64_ps (__mmask8 __U, __m128i __A) {
static __inline__ __m128 __DEFAULT_FN_ATTRS256
_mm256_cvtepu64_ps (__m256i __A) {
return (__m128) __builtin_ia32_cvtuqq2ps256_mask ((__v4di) __A,
(__v4sf) _mm_setzero_ps(),
(__mmask8) -1);
return (__m128)__builtin_convertvector((__v4du)__A, __v4sf);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS256
_mm256_mask_cvtepu64_ps (__m128 __W, __mmask8 __U, __m256i __A) {
return (__m128) __builtin_ia32_cvtuqq2ps256_mask ((__v4di) __A,
(__v4sf) __W,
(__mmask8) __U);
return (__m128)__builtin_ia32_selectps_128((__mmask8)__U,
(__v4sf)_mm256_cvtepu64_ps(__A),
(__v4sf)__W);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS256
_mm256_maskz_cvtepu64_ps (__mmask8 __U, __m256i __A) {
return (__m128) __builtin_ia32_cvtuqq2ps256_mask ((__v4di) __A,
(__v4sf) _mm_setzero_ps(),
(__mmask8) __U);
return (__m128)__builtin_ia32_selectps_128((__mmask8)__U,
(__v4sf)_mm256_cvtepu64_ps(__A),
(__v4sf)_mm_setzero_ps());
}
#define _mm_range_pd(A, B, C) \

16
clang/test/CodeGen/avx512vldq-builtins.c
View File

@ -479,19 +479,21 @@ __m128 test_mm_maskz_cvtepi64_ps(__mmask8 __U, __m128i __A) {
__m128 test_mm256_cvtepi64_ps(__m256i __A) {
// CHECK-LABEL: @test_mm256_cvtepi64_ps
// CHECK: @llvm.x86.avx512.mask.cvtqq2ps.256
// CHECK: sitofp <4 x i64> %{{.*}} to <4 x float>
return _mm256_cvtepi64_ps(__A);
}
__m128 test_mm256_mask_cvtepi64_ps(__m128 __W, __mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_mask_cvtepi64_ps
// CHECK: @llvm.x86.avx512.mask.cvtqq2ps.256
// CHECK: sitofp <4 x i64> %{{.*}} to <4 x float>
// select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
return _mm256_mask_cvtepi64_ps(__W, __U, __A);
}
__m128 test_mm256_maskz_cvtepi64_ps(__mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_maskz_cvtepi64_ps
// CHECK: @llvm.x86.avx512.mask.cvtqq2ps.256
// CHECK: sitofp <4 x i64> %{{.*}} to <4 x float>
// select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
return _mm256_maskz_cvtepi64_ps(__U, __A);
}
@ -699,19 +701,21 @@ __m128 test_mm_maskz_cvtepu64_ps(__mmask8 __U, __m128i __A) {
__m128 test_mm256_cvtepu64_ps(__m256i __A) {
// CHECK-LABEL: @test_mm256_cvtepu64_ps
// CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.256
// CHECK: uitofp <4 x i64> %{{.*}} to <4 x float>
return _mm256_cvtepu64_ps(__A);
}
__m128 test_mm256_mask_cvtepu64_ps(__m128 __W, __mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_mask_cvtepu64_ps
// CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.256
// CHECK: uitofp <4 x i64> %{{.*}} to <4 x float>
// CHECK: select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
return _mm256_mask_cvtepu64_ps(__W, __U, __A);
}
__m128 test_mm256_maskz_cvtepu64_ps(__mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_maskz_cvtepu64_ps
// CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.256
// CHECK: uitofp <4 x i64> %{{.*}} to <4 x float>
// CHECK: select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
return _mm256_maskz_cvtepu64_ps(__U, __A);
}