r228481 - Make the byte-shift SSE intrinsics emit vector shuffles which we know the backend can handle.
Filipe Cabecinhas
me at filcab.net
Fri Feb 6 17:37:09 PST 2015
Author: filcab
Date: Fri Feb 6 19:37:09 2015
New Revision: 228481
URL: http://llvm.org/viewvc/llvm-project?rev=228481&view=rev
Log:
Make the byte-shift SSE intrinsics emit vector shuffles which we know the backend can handle.
Also removed unused builtins.
Original patch by Andrea Di Biagio!
Reviewers: craig.topper, nadav
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D7199
Added:
cfe/trunk/test/CodeGen/sse.c
Modified:
cfe/trunk/include/clang/Basic/BuiltinsX86.def
cfe/trunk/lib/Headers/emmintrin.h
cfe/trunk/test/Sema/x86-builtin-palignr.c
Modified: cfe/trunk/include/clang/Basic/BuiltinsX86.def
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/BuiltinsX86.def?rev=228481&r1=228480&r2=228481&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/BuiltinsX86.def (original)
+++ cfe/trunk/include/clang/Basic/BuiltinsX86.def Fri Feb 6 19:37:09 2015
@@ -316,8 +316,6 @@ BUILTIN(__builtin_ia32_psraw128, "V8sV8s
BUILTIN(__builtin_ia32_psrad128, "V4iV4iV4i", "")
BUILTIN(__builtin_ia32_psrlw128, "V8sV8sV8s", "")
BUILTIN(__builtin_ia32_psrld128, "V4iV4iV4i", "")
-BUILTIN(__builtin_ia32_pslldqi128, "V2LLiV2LLiIi", "")
-BUILTIN(__builtin_ia32_psrldqi128, "V2LLiV2LLiIi", "")
BUILTIN(__builtin_ia32_psrlq128, "V2LLiV2LLiV2LLi", "")
BUILTIN(__builtin_ia32_psllw128, "V8sV8sV8s", "")
BUILTIN(__builtin_ia32_pslld128, "V4iV4iV4i", "")
Modified: cfe/trunk/lib/Headers/emmintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/emmintrin.h?rev=228481&r1=228480&r2=228481&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/emmintrin.h (original)
+++ cfe/trunk/lib/Headers/emmintrin.h Fri Feb 6 19:37:09 2015
@@ -825,11 +825,30 @@ _mm_xor_si128(__m128i __a, __m128i __b)
return __a ^ __b;
}
-#define _mm_slli_si128(a, count) __extension__ ({ \
- _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wshadow\""); \
- __m128i __a = (a); \
- _Pragma("clang diagnostic pop"); \
- (__m128i)__builtin_ia32_pslldqi128(__a, (count)*8); })
+#define _mm_slli_si128(a, imm) __extension__({ \
+ _Pragma("clang diagnostic push") \
+ _Pragma("clang diagnostic ignored \"-Wshadow\""); \
+ __m128i __a = (a); \
+ _Pragma("clang diagnostic pop"); \
+ (__m128i) \
+ __builtin_shufflevector((__v16qi)_mm_setzero_si128(), (__v16qi)__a, \
+ ((imm)&0xF0) ? 0 : 16 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 17 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 18 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 19 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 20 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 21 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 22 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 23 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 24 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 25 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 26 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 27 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 28 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 29 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 30 - ((imm)&0xF), \
+ ((imm)&0xF0) ? 0 : 31 - ((imm)&0xF)); \
+ })
static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
_mm_slli_epi16(__m128i __a, int __count)
@@ -891,12 +910,30 @@ _mm_sra_epi32(__m128i __a, __m128i __cou
return (__m128i)__builtin_ia32_psrad128((__v4si)__a, (__v4si)__count);
}
-
-#define _mm_srli_si128(a, count) __extension__ ({ \
- _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wshadow\""); \
- __m128i __a = (a); \
- _Pragma("clang diagnostic pop"); \
- (__m128i)__builtin_ia32_psrldqi128(__a, (count)*8); })
+#define _mm_srli_si128(a, imm) __extension__({ \
+ _Pragma("clang diagnostic push") \
+ _Pragma("clang diagnostic ignored \"-Wshadow\""); \
+ __m128i __a = (a); \
+ _Pragma("clang diagnostic pop"); \
+ (__m128i) \
+ __builtin_shufflevector((__v16qi)__a, (__v16qi)_mm_setzero_si128(), \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 0, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 1, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 2, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 3, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 4, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 5, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 6, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 7, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 8, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 9, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 10, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 11, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 12, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 13, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 14, \
+ ((imm)&0xF0) ? 16 : ((imm)&0xF) + 15); \
+ })
static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
_mm_srli_epi16(__m128i __a, int __count)
Added: cfe/trunk/test/CodeGen/sse.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/sse.c?rev=228481&view=auto
==============================================================================
--- cfe/trunk/test/CodeGen/sse.c (added)
+++ cfe/trunk/test/CodeGen/sse.c Fri Feb 6 19:37:09 2015
@@ -0,0 +1,40 @@
+// RUN: %clang_cc1 -O3 -ffreestanding -triple x86_64-apple-macosx10.8.0 -target-feature +sse4.1 -emit-llvm %s -o - | FileCheck %s
+
+#include <emmintrin.h>
+
+// Byte-shifts look reversed due to xmm register layout
+__m128 test_mm_slli_si128(__m128 a) {
+ // CHECK-LABEL: @test_mm_slli_si128
+ // CHECK: shufflevector <16 x i8> <{{.*}}, i8 0, i8 0, i8 0, i8 0, i8 0>, <16 x i8> {{.*}}, <16 x i32> <i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26>
+ return _mm_slli_si128(a, 5);
+}
+
+__m128 test_mm_slli_si128_0(__m128 a) {
+ // CHECK-LABEL: @test_mm_slli_si128_0
+ // CHECK-NOT: shufflevector
+ return _mm_slli_si128(a, 0);
+}
+
+__m128 test_mm_slli_si128_16(__m128 a) {
+ // CHECK-LABEL: @test_mm_slli_si128_16
+ // CHECK-NOT: shufflevector
+ return _mm_slli_si128(a, 16);
+}
+
+__m128 test_mm_srli_si128(__m128 a) {
+ // CHECK-LABEL: @test_mm_srli_si128
+ // CHECK: shufflevector <16 x i8> {{.*}}, <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, {{.*}}>, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>
+ return _mm_srli_si128(a, 5);
+}
+
+__m128 test_mm_srli_si128_0(__m128 a) {
+ // CHECK-LABEL: @test_mm_srli_si128_0
+ // CHECK-NOT: shufflevector
+ return _mm_srli_si128(a, 0);
+}
+
+__m128 test_mm_srli_si128_16(__m128 a) {
+ // CHECK-LABEL: @test_mm_srli_si128_16
+ // CHECK-NOT: shufflevector
+ return _mm_srli_si128(a, 16);
+}
Modified: cfe/trunk/test/Sema/x86-builtin-palignr.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/Sema/x86-builtin-palignr.c?rev=228481&r1=228480&r2=228481&view=diff
==============================================================================
--- cfe/trunk/test/Sema/x86-builtin-palignr.c (original)
+++ cfe/trunk/test/Sema/x86-builtin-palignr.c Fri Feb 6 19:37:09 2015
@@ -8,10 +8,3 @@ __m64 test1(__m64 a, __m64 b, int c) {
// be removed when that is fixed.
return _mm_alignr_pi8(a, b, c); // expected-error {{argument to '__builtin_ia32_palignr' must be a constant integer}} expected-error {{incompatible result type}}
}
-
-int test2(int N) {
- __m128i white2;
- white2 = __builtin_ia32_pslldqi128(white2, N); // expected-error {{argument to '__builtin_ia32_pslldqi128' must be a constant integer}}
- return 0;
-}
-
More information about the cfe-commits
mailing list