r272532 - [AVX512] Masked pcmpeqd, pcmpeqq, pcmpgtd, and pcmpgtq don't require avx512bw, just avx512vl.
Craig Topper via cfe-commits
cfe-commits at lists.llvm.org
Sun Jun 12 21:15:11 PDT 2016
Author: ctopper
Date: Sun Jun 12 23:15:11 2016
New Revision: 272532
URL: http://llvm.org/viewvc/llvm-project?rev=272532&view=rev
Log:
[AVX512] Masked pcmpeqd, pcmpeqq, pcmpgtd, and pcmpgtq don't require avx512bw, just avx512vl.
Modified:
cfe/trunk/include/clang/Basic/BuiltinsX86.def
cfe/trunk/lib/Headers/avx512vlintrin.h
cfe/trunk/test/CodeGen/avx512vl-builtins.c
Modified: cfe/trunk/include/clang/Basic/BuiltinsX86.def
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/BuiltinsX86.def?rev=272532&r1=272531&r2=272532&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/BuiltinsX86.def (original)
+++ cfe/trunk/include/clang/Basic/BuiltinsX86.def Sun Jun 12 23:15:11 2016
@@ -939,12 +939,12 @@ TARGET_BUILTIN(__builtin_ia32_pcmpeqq512
TARGET_BUILTIN(__builtin_ia32_pcmpeqw512_mask, "iV32sV32si", "", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_pcmpeqb256_mask, "iV32cV32ci", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pcmpeqd256_mask, "cV8iV8ic", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pcmpeqq256_mask, "cV4LLiV4LLic", "", "avx512vl,avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pcmpeqd256_mask, "cV8iV8ic", "", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pcmpeqq256_mask, "cV4LLiV4LLic", "", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_pcmpeqw256_mask, "sV16sV16ss", "", "avx512vl,avx512bw")
TARGET_BUILTIN(__builtin_ia32_pcmpeqb128_mask, "sV16cV16cs", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pcmpeqd128_mask, "cV4iV4ic", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pcmpeqq128_mask, "cV2LLiV2LLic", "", "avx512vl,avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pcmpeqd128_mask, "cV4iV4ic", "", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pcmpeqq128_mask, "cV2LLiV2LLic", "", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_pcmpeqw128_mask, "cV8sV8sc", "", "avx512vl,avx512bw")
TARGET_BUILTIN(__builtin_ia32_pcmpgtb512_mask, "LLiV64cV64cLLi", "", "avx512bw")
@@ -953,12 +953,12 @@ TARGET_BUILTIN(__builtin_ia32_pcmpgtq512
TARGET_BUILTIN(__builtin_ia32_pcmpgtw512_mask, "iV32sV32si", "", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_pcmpgtb256_mask, "iV32cV32ci", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pcmpgtd256_mask, "cV8iV8ic", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pcmpgtq256_mask, "cV4LLiV4LLic", "", "avx512vl,avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pcmpgtd256_mask, "cV8iV8ic", "", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pcmpgtq256_mask, "cV4LLiV4LLic", "", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_pcmpgtw256_mask, "sV16sV16ss", "", "avx512vl,avx512bw")
TARGET_BUILTIN(__builtin_ia32_pcmpgtb128_mask, "sV16cV16cs", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pcmpgtd128_mask, "cV4iV4ic", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pcmpgtq128_mask, "cV2LLiV2LLic", "", "avx512vl,avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pcmpgtd128_mask, "cV4iV4ic", "", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pcmpgtq128_mask, "cV2LLiV2LLic", "", "avx512vl")
TARGET_BUILTIN(__builtin_ia32_pcmpgtw128_mask, "cV8sV8sc", "", "avx512vl,avx512bw")
TARGET_BUILTIN(__builtin_ia32_cmppd512_mask, "UcV8dV8dIiUcIi", "", "avx512f")
Modified: cfe/trunk/lib/Headers/avx512vlintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/avx512vlintrin.h?rev=272532&r1=272531&r2=272532&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/avx512vlintrin.h (original)
+++ cfe/trunk/lib/Headers/avx512vlintrin.h Sun Jun 12 23:15:11 2016
@@ -29,7 +29,6 @@
#define __AVX512VLINTRIN_H
#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512vl")))
-#define __DEFAULT_FN_ATTRS_BOTH __attribute__((__always_inline__, __nodebug__, __target__("avx512vl, avx512bw")))
static __inline __v2di __DEFAULT_FN_ATTRS
_mm_setzero_di(void) {
@@ -38,13 +37,13 @@ _mm_setzero_di(void) {
/* Integer compare */
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm_cmpeq_epi32_mask(__m128i __a, __m128i __b) {
return (__mmask8)__builtin_ia32_pcmpeqd128_mask((__v4si)__a, (__v4si)__b,
(__mmask8)-1);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm_mask_cmpeq_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
return (__mmask8)__builtin_ia32_pcmpeqd128_mask((__v4si)__a, (__v4si)__b,
__u);
@@ -62,13 +61,13 @@ _mm_mask_cmpeq_epu32_mask(__mmask8 __u,
__u);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm256_cmpeq_epi32_mask(__m256i __a, __m256i __b) {
return (__mmask8)__builtin_ia32_pcmpeqd256_mask((__v8si)__a, (__v8si)__b,
(__mmask8)-1);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm256_mask_cmpeq_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
return (__mmask8)__builtin_ia32_pcmpeqd256_mask((__v8si)__a, (__v8si)__b,
__u);
@@ -86,13 +85,13 @@ _mm256_mask_cmpeq_epu32_mask(__mmask8 __
__u);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm_cmpeq_epi64_mask(__m128i __a, __m128i __b) {
return (__mmask8)__builtin_ia32_pcmpeqq128_mask((__v2di)__a, (__v2di)__b,
(__mmask8)-1);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm_mask_cmpeq_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
return (__mmask8)__builtin_ia32_pcmpeqq128_mask((__v2di)__a, (__v2di)__b,
__u);
@@ -110,13 +109,13 @@ _mm_mask_cmpeq_epu64_mask(__mmask8 __u,
__u);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm256_cmpeq_epi64_mask(__m256i __a, __m256i __b) {
return (__mmask8)__builtin_ia32_pcmpeqq256_mask((__v4di)__a, (__v4di)__b,
(__mmask8)-1);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm256_mask_cmpeq_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
return (__mmask8)__builtin_ia32_pcmpeqq256_mask((__v4di)__a, (__v4di)__b,
__u);
@@ -231,13 +230,13 @@ _mm256_mask_cmpge_epu64_mask(__mmask8 __
__u);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm_cmpgt_epi32_mask(__m128i __a, __m128i __b) {
return (__mmask8)__builtin_ia32_pcmpgtd128_mask((__v4si)__a, (__v4si)__b,
(__mmask8)-1);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm_mask_cmpgt_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
return (__mmask8)__builtin_ia32_pcmpgtd128_mask((__v4si)__a, (__v4si)__b,
__u);
@@ -255,13 +254,13 @@ _mm_mask_cmpgt_epu32_mask(__mmask8 __u,
__u);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm256_cmpgt_epi32_mask(__m256i __a, __m256i __b) {
return (__mmask8)__builtin_ia32_pcmpgtd256_mask((__v8si)__a, (__v8si)__b,
(__mmask8)-1);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm256_mask_cmpgt_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
return (__mmask8)__builtin_ia32_pcmpgtd256_mask((__v8si)__a, (__v8si)__b,
__u);
@@ -279,13 +278,13 @@ _mm256_mask_cmpgt_epu32_mask(__mmask8 __
__u);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm_cmpgt_epi64_mask(__m128i __a, __m128i __b) {
return (__mmask8)__builtin_ia32_pcmpgtq128_mask((__v2di)__a, (__v2di)__b,
(__mmask8)-1);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm_mask_cmpgt_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
return (__mmask8)__builtin_ia32_pcmpgtq128_mask((__v2di)__a, (__v2di)__b,
__u);
@@ -303,13 +302,13 @@ _mm_mask_cmpgt_epu64_mask(__mmask8 __u,
__u);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm256_cmpgt_epi64_mask(__m256i __a, __m256i __b) {
return (__mmask8)__builtin_ia32_pcmpgtq256_mask((__v4di)__a, (__v4di)__b,
(__mmask8)-1);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS_BOTH
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
_mm256_mask_cmpgt_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
return (__mmask8)__builtin_ia32_pcmpgtq256_mask((__v4di)__a, (__v4di)__b,
__u);
@@ -9427,6 +9426,5 @@ _mm256_maskz_cvtps_ph ( __mmask8 __U, __
#undef __DEFAULT_FN_ATTRS
-#undef __DEFAULT_FN_ATTRS_BOTH
#endif /* __AVX512VLINTRIN_H */
Modified: cfe/trunk/test/CodeGen/avx512vl-builtins.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/avx512vl-builtins.c?rev=272532&r1=272531&r2=272532&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/avx512vl-builtins.c (original)
+++ cfe/trunk/test/CodeGen/avx512vl-builtins.c Sun Jun 12 23:15:11 2016
@@ -6738,3 +6738,99 @@ __m128i test_mm256_maskz_cvt_roundps_ph(
// CHECK: @llvm.x86.avx512.mask.vcvtps2ph.256
return _mm256_maskz_cvt_roundps_ph(__U, __A, _MM_FROUND_CUR_DIRECTION);
}
+
+__mmask8 test_mm_cmpeq_epi32_mask(__m128i __a, __m128i __b) {
+ // CHECK-LABEL: @test_mm_cmpeq_epi32_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpeq.d.128
+ return (__mmask8)_mm_cmpeq_epi32_mask(__a, __b);
+}
+
+__mmask8 test_mm_mask_cmpeq_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+ // CHECK-LABEL: @test_mm_mask_cmpeq_epi32_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpeq.d.128
+ return (__mmask8)_mm_mask_cmpeq_epi32_mask(__u, __a, __b);
+}
+
+__mmask8 test_mm_mask_cmpeq_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+ // CHECK-LABEL: @test_mm_mask_cmpeq_epi64_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpeq.q.128
+ return (__mmask8)_mm_mask_cmpeq_epi64_mask(__u, __a, __b);
+}
+
+__mmask8 test_mm_cmpeq_epi64_mask(__m128i __a, __m128i __b) {
+ // CHECK-LABEL: @test_mm_cmpeq_epi64_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpeq.q.128
+ return (__mmask8)_mm_cmpeq_epi64_mask(__a, __b);
+}
+
+__mmask8 test_mm_cmpgt_epi32_mask(__m128i __a, __m128i __b) {
+ // CHECK-LABEL: @test_mm_cmpgt_epi32_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpgt.d.128
+ return (__mmask8)_mm_cmpgt_epi32_mask(__a, __b);
+}
+
+__mmask8 test_mm_mask_cmpgt_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+ // CHECK-LABEL: @test_mm_mask_cmpgt_epi32_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpgt.d.128
+ return (__mmask8)_mm_mask_cmpgt_epi32_mask(__u, __a, __b);
+}
+
+__mmask8 test_mm_mask_cmpgt_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+ // CHECK-LABEL: @test_mm_mask_cmpgt_epi64_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpgt.q.128
+ return (__mmask8)_mm_mask_cmpgt_epi64_mask(__u, __a, __b);
+}
+
+__mmask8 test_mm_cmpgt_epi64_mask(__m128i __a, __m128i __b) {
+ // CHECK-LABEL: @test_mm_cmpgt_epi64_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpgt.q.128
+ return (__mmask8)_mm_cmpgt_epi64_mask(__a, __b);
+}
+
+__mmask8 test_mm256_cmpeq_epi32_mask(__m256i __a, __m256i __b) {
+ // CHECK-LABEL: @test_mm256_cmpeq_epi32_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpeq.d.256
+ return (__mmask8)_mm256_cmpeq_epi32_mask(__a, __b);
+}
+
+__mmask8 test_mm256_mask_cmpeq_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+ // CHECK-LABEL: @test_mm256_mask_cmpeq_epi32_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpeq.d.256
+ return (__mmask8)_mm256_mask_cmpeq_epi32_mask(__u, __a, __b);
+}
+
+__mmask8 test_mm256_mask_cmpeq_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+ // CHECK-LABEL: @test_mm256_mask_cmpeq_epi64_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpeq.q.256
+ return (__mmask8)_mm256_mask_cmpeq_epi64_mask(__u, __a, __b);
+}
+
+__mmask8 test_mm256_cmpeq_epi64_mask(__m256i __a, __m256i __b) {
+ // CHECK-LABEL: @test_mm256_cmpeq_epi64_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpeq.q.256
+ return (__mmask8)_mm256_cmpeq_epi64_mask(__a, __b);
+}
+
+__mmask8 test_mm256_cmpgt_epi32_mask(__m256i __a, __m256i __b) {
+ // CHECK-LABEL: @test_mm256_cmpgt_epi32_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpgt.d.256
+ return (__mmask8)_mm256_cmpgt_epi32_mask(__a, __b);
+}
+
+__mmask8 test_mm256_mask_cmpgt_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+ // CHECK-LABEL: @test_mm256_mask_cmpgt_epi32_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpgt.d.256
+ return (__mmask8)_mm256_mask_cmpgt_epi32_mask(__u, __a, __b);
+}
+
+__mmask8 test_mm256_mask_cmpgt_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+ // CHECK-LABEL: @test_mm256_mask_cmpgt_epi64_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpgt.q.256
+ return (__mmask8)_mm256_mask_cmpgt_epi64_mask(__u, __a, __b);
+}
+
+__mmask8 test_mm256_cmpgt_epi64_mask(__m256i __a, __m256i __b) {
+ // CHECK-LABEL: @test_mm256_cmpgt_epi64_mask
+ // CHECK: @llvm.x86.avx512.mask.pcmpgt.q.256
+ return (__mmask8)_mm256_cmpgt_epi64_mask(__a, __b);
+}
More information about the cfe-commits
mailing list