[X86][SSE] Replace (V)CVTTPS2DQ and VCVTTPD2DQ truncating (round to zero) f32/f64...

The 'cvtt' truncation (round to zero) conversions can be safely represented as generic __builtin_convertvector (fptosi) calls instead of x86 intrinsics. We already do this (implicitly) for the scalar equivalents.

Note: I looked at updating _mm_cvttpd_epi32 as well but this still requires a lot more backend work to correctly lower (both for debug and optimized builds).

Differential Revision: http://reviews.llvm.org/D20859

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@271436 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index b4c44437d907cfb34b6a323383f02400f5fbf23f..7f95c3f1f216a005571ec16277e0cc45c108aded 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -339,7 +339,6 @@ TARGET_BUILTIN(__builtin_ia32_cvttpd2dq, "V4iV2d", "", "sse2")
 TARGET_BUILTIN(__builtin_ia32_cvtsd2si, "iV2d", "", "sse2")
 TARGET_BUILTIN(__builtin_ia32_cvtsd2si64, "LLiV2d", "", "sse2")
 TARGET_BUILTIN(__builtin_ia32_cvtps2dq, "V4iV4f", "", "sse2")
-TARGET_BUILTIN(__builtin_ia32_cvttps2dq, "V4iV4f", "", "sse2")
 TARGET_BUILTIN(__builtin_ia32_clflush, "vvC*", "", "sse2")
 TARGET_BUILTIN(__builtin_ia32_lfence, "v", "", "sse2")
 TARGET_BUILTIN(__builtin_ia32_mfence, "v", "", "sse2")
@@ -462,9 +461,7 @@ TARGET_BUILTIN(__builtin_ia32_cmpps256, "V8fV8fV8fIc", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_cvtdq2ps256, "V8fV8i", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_cvtpd2ps256, "V4fV4d", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_cvtps2dq256, "V8iV8f", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_cvttpd2dq256, "V4iV4d", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_cvtpd2dq256, "V4iV4d", "", "avx")
-TARGET_BUILTIN(__builtin_ia32_cvttps2dq256, "V8iV8f", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_vperm2f128_pd256, "V4dV4dV4dIc", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_vperm2f128_ps256, "V8fV8fV8fIc", "", "avx")
 TARGET_BUILTIN(__builtin_ia32_vperm2f128_si256, "V8iV8iV8iIc", "", "avx")

diff --git a/lib/Headers/avxintrin.h b/lib/Headers/avxintrin.h
index 6a7a4b903a6f87d4a4e201c88becfc82cdc82fc1..e46f8f4ac41d6e100ffb1c65f2a2f3bba7ff985e 100644 (file)
--- a/lib/Headers/avxintrin.h
+++ b/lib/Headers/avxintrin.h
@@ -2108,7 +2108,7 @@ _mm256_cvtps_pd(__m128 __a)
 static __inline __m128i __DEFAULT_FN_ATTRS
 _mm256_cvttpd_epi32(__m256d __a)
 {
-  return (__m128i)__builtin_ia32_cvttpd2dq256((__v4df) __a);
+  return (__m128i)__builtin_convertvector((__v4df) __a, __v4si);
 }
 
 static __inline __m128i __DEFAULT_FN_ATTRS
@@ -2120,7 +2120,7 @@ _mm256_cvtpd_epi32(__m256d __a)
 static __inline __m256i __DEFAULT_FN_ATTRS
 _mm256_cvttps_epi32(__m256 __a)
 {
-  return (__m256i)__builtin_ia32_cvttps2dq256((__v8sf) __a);
+  return (__m256i)__builtin_convertvector((__v8sf) __a, __v8si);
 }
 
 static __inline double __DEFAULT_FN_ATTRS

diff --git a/lib/Headers/emmintrin.h b/lib/Headers/emmintrin.h
index 08ee06b1c61634b3ab85a0cabe45b4fcaa9ed01b..04cca74a425c254cdb75cb5159f5629c37f91816 100644 (file)
--- a/lib/Headers/emmintrin.h
+++ b/lib/Headers/emmintrin.h
@@ -1744,7 +1744,7 @@ _mm_cvtps_epi32(__m128 __a)
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_cvttps_epi32(__m128 __a)
 {
-  return (__m128i)__builtin_ia32_cvttps2dq((__v4sf)__a);
+  return (__m128i)__builtin_convertvector((__v4sf)__a, __v4si);
 }
 
 /// \brief Returns a vector of [4 x i32] where the lowest element is the input

diff --git a/test/CodeGen/avx-builtins.c b/test/CodeGen/avx-builtins.c
index c1b8cf2c919aa11074e10cf74274abff319479f9..f1f211e54d33a2b3473332b1774fa396fe785f7b 100644 (file)
--- a/test/CodeGen/avx-builtins.c
+++ b/test/CodeGen/avx-builtins.c
@@ -286,13 +286,13 @@ __m256d test_mm256_cvtps_pd(__m128 A) {
 
 __m128i test_mm256_cvttpd_epi32(__m256d A) {
   // CHECK-LABEL: test_mm256_cvttpd_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx.cvtt.pd2dq.256(<4 x double> %{{.*}})
+  // CHECK: fptosi <4 x double> %{{.*}} to <4 x i32>
   return _mm256_cvttpd_epi32(A);
 }
 
 __m256i test_mm256_cvttps_epi32(__m256 A) {
   // CHECK-LABEL: test_mm256_cvttps_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx.cvtt.ps2dq.256(<8 x float> %{{.*}})
+  // CHECK: fptosi <8 x float> %{{.*}} to <8 x i32>
   return _mm256_cvttps_epi32(A);
 }
 

diff --git a/test/CodeGen/builtins-x86.c b/test/CodeGen/builtins-x86.c
index 6bfff11b781e1db5d45dc5d1b64d866cbcb10daf..24e491c026226860ecacf5f27b71b58d76065c97 100644 (file)
--- a/test/CodeGen/builtins-x86.c
+++ b/test/CodeGen/builtins-x86.c
@@ -335,7 +335,6 @@ void f0() {
   tmp_LLi = __builtin_ia32_cvtsd2si64(tmp_V2d);
 #endif
   tmp_V4i = __builtin_ia32_cvtps2dq(tmp_V4f);
-  tmp_V4i = __builtin_ia32_cvttps2dq(tmp_V4f);
   (void) __builtin_ia32_clflush(tmp_vCp);
   (void) __builtin_ia32_lfence();
   (void) __builtin_ia32_mfence();
@@ -415,9 +414,7 @@ void f0() {
   tmp_V8f = __builtin_ia32_cvtdq2ps256(tmp_V8i);
   tmp_V4f = __builtin_ia32_cvtpd2ps256(tmp_V4d);
   tmp_V8i = __builtin_ia32_cvtps2dq256(tmp_V8f);
-  tmp_V4i = __builtin_ia32_cvttpd2dq256(tmp_V4d);
   tmp_V4i = __builtin_ia32_cvtpd2dq256(tmp_V4d);
-  tmp_V8i = __builtin_ia32_cvttps2dq256(tmp_V8f);
   tmp_V4d = __builtin_ia32_vperm2f128_pd256(tmp_V4d, tmp_V4d, 0x7);
   tmp_V8f = __builtin_ia32_vperm2f128_ps256(tmp_V8f, tmp_V8f, 0x7);
   tmp_V8i = __builtin_ia32_vperm2f128_si256(tmp_V8i, tmp_V8i, 0x7);

diff --git a/test/CodeGen/sse2-builtins.c b/test/CodeGen/sse2-builtins.c
index 7d2cbc20e3c64e5a8b46cf1dc5bb07af965a41b9..f08feb0e3af1cf6df46d8142866cdb584ae21539 100644 (file)
--- a/test/CodeGen/sse2-builtins.c
+++ b/test/CodeGen/sse2-builtins.c
@@ -533,7 +533,7 @@ __m128i test_mm_cvttpd_epi32(__m128d A) {
 
 __m128i test_mm_cvttps_epi32(__m128 A) {
   // CHECK-LABEL: test_mm_cvttps_epi32
-  // CHECK: call <4 x i32> @llvm.x86.sse2.cvttps2dq(<4 x float> %{{.*}})
+  // CHECK: fptosi <4 x float> %{{.*}} to <4 x i32>
   return _mm_cvttps_epi32(A);
 }