[X86] Use __builtin_convertvector to replace some of the avx512 truncate builtins.

As long as the destination type is a 256 or 128 bit vector with the same number of elements we can use __builtin_convertvector to directly generate trunc IR instruction which will be handled natively by the backend.

Differential Revision: https://reviews.llvm.org/D46742

llvm-svn: 332266

4 files changed, 56 insertions, 66 deletions

diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index f5ff5d3c10a..499bb8a6906 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -1157,23 +1157,21 @@ _mm512_maskz_cvtusepi16_epi8 (__mmask32 __M, __m512i __A) {
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_cvtepi16_epi8 (__m512i __A) {
-  return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
-              (__v32qi) _mm256_setzero_si256(),
-              (__mmask32) -1);
+  return (__m256i)__builtin_convertvector((__v32hi)__A, __v32qi);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi16_epi8 (__m256i __O, __mmask32 __M, __m512i __A) {
-  return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
-              (__v32qi) __O,
-              __M);
+  return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
+                                             (__v32qi)_mm512_cvtepi16_epi8(__A),
+                                             (__v32qi)__O);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi16_epi8 (__mmask32 __M, __m512i __A) {
-  return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
-              (__v32qi) _mm256_setzero_si256(),
-              __M);
+  return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
+                                             (__v32qi)_mm512_cvtepi16_epi8(__A),
+                                             (__v32qi)_mm256_setzero_si256());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index e891b6c2d1c..24a4e9fef24 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -7601,24 +7601,23 @@ _mm512_mask_cvtusepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A)
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_cvtepi32_epi8 (__m512i __A)
 {
-  return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A,
-              (__v16qi) _mm_undefined_si128 (),
-              (__mmask16) -1);
+  return (__m128i)__builtin_convertvector((__v16si)__A, __v16qi);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi32_epi8 (__m128i __O, __mmask16 __M, __m512i __A)
 {
-  return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A,
-              (__v16qi) __O, __M);
+  return (__m128i)__builtin_ia32_selectb_128((__mmask8)__M,
+                                             (__v16qi)_mm512_cvtepi32_epi8(__A),
+                                             (__v16qi)__O);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi32_epi8 (__mmask16 __M, __m512i __A)
 {
-  return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A,
-              (__v16qi) _mm_setzero_si128 (),
-              __M);
+  return (__m128i)__builtin_ia32_selectb_128((__mmask8)__M,
+                                             (__v16qi)_mm512_cvtepi32_epi8(__A),
+                                             (__v16qi)_mm_setzero_si128());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
@@ -7630,24 +7629,23 @@ _mm512_mask_cvtepi32_storeu_epi8 (void * __P, __mmask16 __M, __m512i __A)
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_cvtepi32_epi16 (__m512i __A)
 {
-  return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A,
-              (__v16hi) _mm256_undefined_si256 (),
-              (__mmask16) -1);
+  return (__m256i)__builtin_convertvector((__v16si)__A, __v16hi);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi32_epi16 (__m256i __O, __mmask16 __M, __m512i __A)
 {
-  return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A,
-              (__v16hi) __O, __M);
+  return (__m256i)__builtin_ia32_selectw_256((__mmask16)__M,
+                                             (__v16hi)_mm512_cvtepi32_epi16(__A),
+                                             (__v16hi)__O);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi32_epi16 (__mmask16 __M, __m512i __A)
 {
-  return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A,
-              (__v16hi) _mm256_setzero_si256 (),
-              __M);
+  return (__m256i)__builtin_ia32_selectw_256((__mmask16)__M,
+                                             (__v16hi)_mm512_cvtepi32_epi16(__A),
+                                             (__v16hi)_mm256_setzero_si256());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
@@ -7688,24 +7686,23 @@ _mm512_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m512i __A)
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_cvtepi64_epi32 (__m512i __A)
 {
-  return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A,
-              (__v8si) _mm256_undefined_si256 (),
-              (__mmask8) -1);
+  return (__m256i)__builtin_convertvector((__v8di) __A, __v8si);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi64_epi32 (__m256i __O, __mmask8 __M, __m512i __A)
 {
-  return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A,
-              (__v8si) __O, __M);
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M,
+                                             (__v8si)_mm512_cvtepi64_epi32(__A),
+                                             (__v8si)__O);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi64_epi32 (__mmask8 __M, __m512i __A)
 {
-  return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A,
-              (__v8si) _mm256_setzero_si256 (),
-              __M);
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M,
+                                             (__v8si)_mm512_cvtepi64_epi32(__A),
+                                             (__v8si)_mm256_setzero_si256());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
@@ -7717,24 +7714,23 @@ _mm512_mask_cvtepi64_storeu_epi32 (void* __P, __mmask8 __M, __m512i __A)
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_cvtepi64_epi16 (__m512i __A)
 {
-  return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A,
-              (__v8hi) _mm_undefined_si128 (),
-              (__mmask8) -1);
+  return (__m128i)__builtin_convertvector((__v8di)__A, __v8hi);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi64_epi16 (__m128i __O, __mmask8 __M, __m512i __A)
 {
-  return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A,
-              (__v8hi) __O, __M);
+  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__M,
+                                             (__v8hi)_mm512_cvtepi64_epi16(__A),
+                                             (__v8hi)__O);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi64_epi16 (__mmask8 __M, __m512i __A)
 {
-  return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A,
-              (__v8hi) _mm_setzero_si128 (),
-              __M);
+  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__M,
+                                             (__v8hi)_mm512_cvtepi64_epi16(__A),
+                                             (__v8hi)_mm_setzero_si128());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avx512vlbwintrin.h b/clang/lib/Headers/avx512vlbwintrin.h
index e940e2b6853..6c5131ccba9 100644
--- a/clang/lib/Headers/avx512vlbwintrin.h
+++ b/clang/lib/Headers/avx512vlbwintrin.h
@@ -1551,23 +1551,21 @@ _mm_mask_cvtusepi16_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_cvtepi16_epi8 (__m256i __A) {
-  return (__m128i) __builtin_ia32_pmovwb256_mask ((__v16hi) __A,
-               (__v16qi) _mm_setzero_si128(),
-               (__mmask16) -1);
+  return (__m128i)__builtin_convertvector((__v16hi) __A, __v16qi);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_mask_cvtepi16_epi8 (__m128i __O, __mmask16 __M, __m256i __A) {
-  return (__m128i) __builtin_ia32_pmovwb256_mask ((__v16hi) __A,
-               (__v16qi) __O,
-               __M);
+  return (__m128i)__builtin_ia32_selectb_128((__mmask8)__M,
+                                             (__v16qi)_mm256_cvtepi16_epi8(__A),
+                                             (__v16qi)__O);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_maskz_cvtepi16_epi8 (__mmask16 __M, __m256i __A) {
-  return (__m128i) __builtin_ia32_pmovwb256_mask ((__v16hi) __A,
-               (__v16qi) _mm_setzero_si128(),
-               __M);
+  return (__m128i)__builtin_ia32_selectb_128((__mmask8)__M,
+                                             (__v16qi)_mm256_cvtepi16_epi8(__A),
+                                             (__v16qi)_mm_setzero_si128());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
index 2581c05c5d8..07429953f91 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -7627,24 +7627,23 @@ _mm_mask_cvtepi32_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_cvtepi32_epi16 (__m256i __A)
 {
-  return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A,
-              (__v8hi)_mm_setzero_si128 (),
-              (__mmask8) -1);
+  return (__m128i)__builtin_convertvector((__v8si)__A, __v8hi);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_mask_cvtepi32_epi16 (__m128i __O, __mmask8 __M, __m256i __A)
 {
-  return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A,
-              (__v8hi) __O, __M);
+  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__M,
+                                             (__v8hi)_mm256_cvtepi32_epi16(__A),
+                                             (__v8hi)__O);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_maskz_cvtepi32_epi16 (__mmask8 __M, __m256i __A)
 {
-  return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A,
-              (__v8hi) _mm_setzero_si128 (),
-              __M);
+  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__M,
+                                             (__v8hi)_mm256_cvtepi32_epi16(__A),
+                                             (__v8hi)_mm_setzero_si128());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
@@ -7743,24 +7742,23 @@ _mm_mask_cvtepi64_storeu_epi32 (void * __P, __mmask8 __M, __m128i __A)
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_cvtepi64_epi32 (__m256i __A)
 {
-  return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A,
-              (__v4si) _mm_undefined_si128(),
-              (__mmask8) -1);
+  return (__m128i)__builtin_convertvector((__v4di)__A, __v4si);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_mask_cvtepi64_epi32 (__m128i __O, __mmask8 __M, __m256i __A)
 {
-  return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A,
-              (__v4si) __O, __M);
+  return (__m128i)__builtin_ia32_selectd_128((__mmask8)__M,
+                                             (__v4si)_mm256_cvtepi64_epi32(__A),
+                                             (__v4si)__O);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_maskz_cvtepi64_epi32 (__mmask8 __M, __m256i __A)
 {
-  return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A,
-              (__v4si) _mm_setzero_si128 (),
-              __M);
+  return (__m128i)__builtin_ia32_selectd_128((__mmask8)__M,
+                                             (__v4si)_mm256_cvtepi64_epi32(__A),
+                                             (__v4si)_mm_setzero_si128());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS