[X86] replace vextractf128 intrinsics with generic shuffles

[spatel]

Summary

This is very much like D8088 (checked in here: http://llvm.org/viewvc/llvm-project?rev=231794&view=rev ).

Now that we've replaced the vinsertf128 intrinsics, do the same for their extract twins.

[andreadb]

LGTM.

[spatel]

Thanks, Andrea.

I checked in the LLVM side of this patch at r232047 without realizing there was a dependency on this patch. Sorry about the build-bot breakage!

[spatel]

In D8275#139797, @spatel wrote:

I checked in the LLVM side of this patch at r232047 without realizing there was a dependency on this patch. Sorry about the build-bot breakage!

Got my patches mixed up. That should have been r232045:
http://llvm.org/viewvc/llvm-project?view=revision&revision=232045

[craig.topper]

lib/Headers/avxintrin.h

1184 ↗

(On Diff #21787)

Not sure if its safe to access V twice. It could be a function call that we maybe shouldn't invoke twice. May need a temporary here even though it might fail Wshadow

Or maybe just feed an appropriate zero vector to the second shuffle argument?

[spatel]

lib/Headers/avxintrin.h

1184 ↗

(On Diff #21787)

Nice catch! I think the zero vector implementation should be safe. Checked in at r232061.

While we're here, do you see a problem with casting the 32-bit / packed single case to use 64-bit values instead? If we do it that way, the masks would be identical for all cases. And then we could just define the 2nd and 3rd macros in terms of the 1st?

cfe/trunk/include/clang/Basic/BuiltinsX86.def

 BUILTIN(__builtin_ia32_vpermilvarps, "V4fV4fV4i", "")
 BUILTIN(__builtin_ia32_vpermilvarpd256, "V4dV4dV4LLi", "")
 BUILTIN(__builtin_ia32_vpermilvarps256, "V8fV8fV8i", "")
 BUILTIN(__builtin_ia32_blendvpd256, "V4dV4dV4dV4d", "")
 BUILTIN(__builtin_ia32_blendvps256, "V8fV8fV8fV8f", "")
 BUILTIN(__builtin_ia32_dpps256, "V8fV8fV8fIc", "")
 BUILTIN(__builtin_ia32_cmppd256, "V4dV4dV4dIc", "")
 BUILTIN(__builtin_ia32_cmpps256, "V8fV8fV8fIc", "")
-BUILTIN(__builtin_ia32_vextractf128_pd256, "V2dV4dIc", "")
-BUILTIN(__builtin_ia32_vextractf128_ps256, "V4fV8fIc", "")
-BUILTIN(__builtin_ia32_vextractf128_si256, "V4iV8iIc", "")
 BUILTIN(__builtin_ia32_cvtdq2pd256, "V4dV4i", "")
 BUILTIN(__builtin_ia32_cvtdq2ps256, "V8fV8i", "")
 BUILTIN(__builtin_ia32_cvtpd2ps256, "V4fV4d", "")
 BUILTIN(__builtin_ia32_cvtps2dq256, "V8iV8f", "")
 BUILTIN(__builtin_ia32_cvtps2pd256, "V4dV4f", "")
 BUILTIN(__builtin_ia32_cvttpd2dq256, "V4iV4d", "")
 BUILTIN(__builtin_ia32_cvtpd2dq256, "V4iV4d", "")
 BUILTIN(__builtin_ia32_cvttps2dq256, "V8iV8f", "")

cfe/trunk/lib/Headers/avxintrin.h

#define _mm_cmp_sd(a, b, c) __extension__ ({ \
   __m128d __b = (b); \
   (__m128d)__builtin_ia32_cmpsd((__v2df)__a, (__v2df)__b, (c)); })
 
 #define _mm_cmp_ss(a, b, c) __extension__ ({ \
   __m128 __a = (a); \
   __m128 __b = (b); \
   (__m128)__builtin_ia32_cmpss((__v4sf)__a, (__v4sf)__b, (c)); })
 
-/* Vector extract */
-#define _mm256_extractf128_pd(A, O) __extension__ ({ \
-  __m256d __A = (A); \
-  (__m128d)__builtin_ia32_vextractf128_pd256((__v4df)__A, (O)); })
-
-#define _mm256_extractf128_ps(A, O) __extension__ ({ \
-  __m256 __A = (A); \
-  (__m128)__builtin_ia32_vextractf128_ps256((__v8sf)__A, (O)); })
-
-#define _mm256_extractf128_si256(A, O) __extension__ ({ \
-  __m256i __A = (A); \
-  (__m128i)__builtin_ia32_vextractf128_si256((__v8si)__A, (O)); })
-
 static __inline int __attribute__((__always_inline__, __nodebug__))
 _mm256_extract_epi32(__m256i __a, const int __imm)
 {
   __v8si __b = (__v8si)__a;
   return __b[__imm & 7];
 }
 
 static __inline int __attribute__((__always_inline__, __nodebug__))
#define _mm256_insertf128_si256(V1, V2, M) __extension__ ({ \
   (__m256i)__builtin_shufflevector( \
     (__v4di)(V1), \
     (__v4di)_mm256_castsi128_si256((__m128i)(V2)), \
     (((M) & 1) ? 0 : 4), \
     (((M) & 1) ? 1 : 5), \
     (((M) & 1) ? 4 : 2), \
     (((M) & 1) ? 5 : 3) );})
 
+/* 
+   Vector extract.
+   We use macros rather than inlines because we only want to accept
+   invocations where the immediate M is a constant expression.
+*/
+#define _mm256_extractf128_ps(V, M) __extension__ ({ \
+  (__m128)__builtin_shufflevector( \
+    (__v8sf)(V), \
+    (__v8sf)(V), \
+    (((M) & 1) ? 4 : 0), \
+    (((M) & 1) ? 5 : 1), \
+    (((M) & 1) ? 6 : 2), \
+    (((M) & 1) ? 7 : 3) );})
+
+#define _mm256_extractf128_pd(V, M) __extension__ ({ \
+  (__m128d)__builtin_shufflevector( \
+    (__v4df)(V), \
+    (__v4df)(V), \
+    (((M) & 1) ? 2 : 0), \
+    (((M) & 1) ? 3 : 1) );})
+
+#define _mm256_extractf128_si256(V, M) __extension__ ({ \
+  (__m128i)__builtin_shufflevector( \
+    (__v4di)(V), \
+    (__v4di)(V), \
+    (((M) & 1) ? 2 : 0), \
+    (((M) & 1) ? 3 : 1) );})
+
 /* SIMD load ops (unaligned) */
 static __inline __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_loadu2_m128(float const *__addr_hi, float const *__addr_lo)
 {
   struct __loadu_ps {
     __m128 __v;
   } __attribute__((__packed__, __may_alias__));
 

cfe/trunk/lib/Sema/SemaChecking.cpp

bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   return SemaBuiltinConstantArgRange(TheCall, i, l, u);
 }
 
 bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   unsigned i = 0, l = 0, u = 0;
   switch (BuiltinID) {
   default: return false;
   case X86::BI_mm_prefetch: i = 1; l = 0; u = 3; break;
-  case X86::BI__builtin_ia32_vextractf128_pd256:
-  case X86::BI__builtin_ia32_vextractf128_ps256:
-  case X86::BI__builtin_ia32_vextractf128_si256:
   case X86::BI__builtin_ia32_extract128i256: i = 1, l = 0, u = 1; break;
   case X86::BI__builtin_ia32_insert128i256: i = 2, l = 0; u = 1; break;
   case X86::BI__builtin_ia32_sha1rnds4: i = 2, l = 0; u = 3; break;
   case X86::BI__builtin_ia32_vpermil2pd:
   case X86::BI__builtin_ia32_vpermil2pd256:
   case X86::BI__builtin_ia32_vpermil2ps:
   case X86::BI__builtin_ia32_vpermil2ps256: i = 3, l = 0; u = 3; break;
   case X86::BI__builtin_ia32_cmpb128_mask:

cfe/trunk/test/CodeGen/avx-shuffle-builtins.c

test_mm256_broadcast_ss(float const *__a) {
   // CHECK: insertelement <8 x float> {{.*}}, i32 5
   // CHECK: insertelement <8 x float> {{.*}}, i32 6
   // CHECK: insertelement <8 x float> {{.*}}, i32 7
   return _mm256_broadcast_ss(__a);
 }
 
 // Make sure we have the correct mask for each insertf128 case.
 
-__m256d test_mm256_insertf128_ps_0(__m256 a, __m128 b) {
+__m256 test_mm256_insertf128_ps_0(__m256 a, __m128 b) {
   // CHECK-LABEL: @test_mm256_insertf128_ps_0
   // CHECK: shufflevector{{.*}}<i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7>
   return _mm256_insertf128_ps(a, b, 0);
 }
 
 __m256d test_mm256_insertf128_pd_0(__m256d a, __m128d b) {
   // CHECK-LABEL: @test_mm256_insertf128_pd_0
   // CHECK: shufflevector{{.*}}<i32 4, i32 5, i32 2, i32 3>
   return _mm256_insertf128_pd(a, b, 0);
 }
 
-__m256d test_mm256_insertf128_si256_0(__m256i a, __m128i b) {
+__m256i test_mm256_insertf128_si256_0(__m256i a, __m128i b) {
   // CHECK-LABEL: @test_mm256_insertf128_si256_0
   // CHECK: shufflevector{{.*}}<i32 4, i32 5, i32 2, i32 3>
   return _mm256_insertf128_si256(a, b, 0);
 }
 
-__m256d test_mm256_insertf128_ps_1(__m256 a, __m128 b) {
+__m256 test_mm256_insertf128_ps_1(__m256 a, __m128 b) {
   // CHECK-LABEL: @test_mm256_insertf128_ps_1
   // CHECK: shufflevector{{.*}}<i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
   return _mm256_insertf128_ps(a, b, 1);
 }
 
 __m256d test_mm256_insertf128_pd_1(__m256d a, __m128d b) {
   // CHECK-LABEL: @test_mm256_insertf128_pd_1
   // CHECK: shufflevector{{.*}}<i32 0, i32 1, i32 4, i32 5>
   return _mm256_insertf128_pd(a, b, 1);
 }
 
-__m256d test_mm256_insertf128_si256_1(__m256i a, __m128i b) {
+__m256i test_mm256_insertf128_si256_1(__m256i a, __m128i b) {
   // CHECK-LABEL: @test_mm256_insertf128_si256_1
   // CHECK: shufflevector{{.*}}<i32 0, i32 1, i32 4, i32 5>
   return _mm256_insertf128_si256(a, b, 1);
 }
 
+// Make sure we have the correct mask for each extractf128 case.
+
+__m128 test_mm256_extractf128_ps_0(__m256 a) {
+  // CHECK-LABEL: @test_mm256_extractf128_ps_0
+  // CHECK: shufflevector{{.*}}<i32 0, i32 1, i32 2, i32 3>
+  return _mm256_extractf128_ps(a, 0);
+}
+
+__m128d test_mm256_extractf128_pd_0(__m256d a) {
+  // CHECK-LABEL: @test_mm256_extractf128_pd_0
+  // CHECK: shufflevector{{.*}}<i32 0, i32 1>
+  return _mm256_extractf128_pd(a, 0);
+}
+
+__m128i test_mm256_extractf128_si256_0(__m256i a) {
+  // CHECK-LABEL: @test_mm256_extractf128_si256_0
+  // CHECK: shufflevector{{.*}}<i32 0, i32 1>
+  return _mm256_extractf128_si256(a, 0);
+}
+
+__m128 test_mm256_extractf128_ps_1(__m256 a) {
+  // CHECK-LABEL: @test_mm256_extractf128_ps_1
+  // CHECK: shufflevector{{.*}}<i32 4, i32 5, i32 6, i32 7>
+  return _mm256_extractf128_ps(a, 1);
+}
+
+__m128d test_mm256_extractf128_pd_1(__m256d a) {
+  // CHECK-LABEL: @test_mm256_extractf128_pd_1
+  // CHECK: shufflevector{{.*}}<i32 2, i32 3>
+  return _mm256_extractf128_pd(a, 1);
+}
+
+__m128i test_mm256_extractf128_si256_1(__m256i a) {
+  // CHECK-LABEL: @test_mm256_extractf128_si256_1
+  // CHECK: shufflevector{{.*}}<i32 2, i32 3>
+  return _mm256_extractf128_si256(a, 1);
+}
+

cfe/trunk/test/CodeGen/builtins-x86.c

#endif
   tmp_V4f = __builtin_ia32_vpermilvarps(tmp_V4f, tmp_V4i);
   tmp_V4d = __builtin_ia32_vpermilvarpd256(tmp_V4d, tmp_V4LLi);
   tmp_V8f = __builtin_ia32_vpermilvarps256(tmp_V8f, tmp_V8i);
   tmp_V4d = __builtin_ia32_blendvpd256(tmp_V4d, tmp_V4d, tmp_V4d);
   tmp_V8f = __builtin_ia32_blendvps256(tmp_V8f, tmp_V8f, tmp_V8f);
   tmp_V8f = __builtin_ia32_dpps256(tmp_V8f, tmp_V8f, 0x7);
   tmp_V4d = __builtin_ia32_cmppd256(tmp_V4d, tmp_V4d, 0);
   tmp_V8f = __builtin_ia32_cmpps256(tmp_V8f, tmp_V8f, 0);
-  tmp_V2d = __builtin_ia32_vextractf128_pd256(tmp_V4d, 0x1);
-  tmp_V4f = __builtin_ia32_vextractf128_ps256(tmp_V8f, 0x1);
-  tmp_V4i = __builtin_ia32_vextractf128_si256(tmp_V8i, 0x1);
   tmp_V4d = __builtin_ia32_cvtdq2pd256(tmp_V4i);
   tmp_V8f = __builtin_ia32_cvtdq2ps256(tmp_V8i);
   tmp_V4f = __builtin_ia32_cvtpd2ps256(tmp_V4d);
   tmp_V8i = __builtin_ia32_cvtps2dq256(tmp_V8f);
   tmp_V4d = __builtin_ia32_cvtps2pd256(tmp_V4f);
   tmp_V4i = __builtin_ia32_cvttpd2dq256(tmp_V4d);
   tmp_V4i = __builtin_ia32_cvtpd2dq256(tmp_V4d);
   tmp_V8i = __builtin_ia32_cvttps2dq256(tmp_V8f);