[X86] Improve the selection of SSE3/AVX addsub instructions.

This patch teaches the backend how to canonicalize a shuffle vectors
according to the rule:

 - (shuffle (FADD A, B), (FSUB A, B), Mask) ->
       (shuffle (FSUB A, -B), (FADD A, -B), Mask)

Where 'Mask' is:
  <0,5,2,7>            ;; for v4f32 and v4f64 shuffles.
  <0,3>                ;; for v2f64 shuffles.
  <0,9,2,11,4,13,6,15> ;; for v8f32 shuffles.

In general, ISel only knows how to pattern-match a canonical
'fadd + fsub + blendi' dag node sequence into an ADDSUB instruction.

This new rule allows to convert a non-canonical dag sequence into a
canonical one that will be matched by a single ADDSUB at ISel stage.

The idea of converting a non-canonical ADDSUB into a canonical one by
swapping the first two operands of the shuffle, and then negating the
second operand of the FADD and FSUB, was originally proposed by Hal Finkel.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211771 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 155 insertions, 2 deletions

diff --git a/test/CodeGen/X86/sse3-avx-addsub.ll b/test/CodeGen/X86/sse3-avx-addsub.ll
index d7874e4145..8b6674312b 100644
--- a/test/CodeGen/X86/sse3-avx-addsub.ll
+++ b/test/CodeGen/X86/sse3-avx-addsub.ll
@@ -3,7 +3,7 @@
 
 ; Test ADDSUB ISel patterns.
 
-; All the functions below are obtained from the following source:
+; Functions below are obtained from the following source:
 ;
 ; typedef double double2 __attribute__((ext_vector_type(2)));
 ; typedef double double4 __attribute__((ext_vector_type(4)));
@@ -19,7 +19,7 @@
 ; float8 test2(float8 A, float8 B) {
 ;   float8 X = A - B;
 ;   float8 Y = A + B;
-;   return (float8){X[0], Y[1], X[2], Y[3], X[4], Y[5], X[6], [7]};
+;   return (float8){X[0], Y[1], X[2], Y[3], X[4], Y[5], X[6], Y[7]};
 ; }
 ;
 ; double4 test3(double4 A, double4 B) {
@@ -141,3 +141,156 @@ define <2 x double> @test4b(<2 x double> %A, <2 x double>* %B) {
 ; AVX: vaddsubpd
 ; CHECK-NEXT: ret
 
+; Functions below are obtained from the following source:
+;
+; float4 test1(float4 A, float4 B) {
+;   float4 X = A + B;
+;   float4 Y = A - B;
+;   return (float4){X[0], Y[1], X[2], Y[3]};
+; }
+;
+; float8 test2(float8 A, float8 B) {
+;   float8 X = A + B;
+;   float8 Y = A - B;
+;   return (float8){X[0], Y[1], X[2], Y[3], X[4], Y[5], X[6], Y[7]};
+; }
+;
+; double4 test3(double4 A, double4 B) {
+;   double4 X = A + B;
+;   double4 Y = A - B;
+;   return (double4){X[0], Y[1], X[2], Y[3]};
+; }
+;
+; double2 test4(double2 A, double2 B) {
+;   double2 X = A + B;
+;   double2 Y = A - B;
+;   return (double2){X[0], Y[1]};
+; }
+
+define <4 x float> @test5(<4 x float> %A, <4 x float> %B) {
+  %sub = fsub <4 x float> %A, %B
+  %add = fadd <4 x float> %A, %B
+  %vecinit6 = shufflevector <4 x float> %add, <4 x float> %sub, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %vecinit6
+}
+; CHECK-LABEL: test5
+; SSE: xorps
+; SSE-NEXT: addsubps
+; AVX: vxorps
+; AVX-NEXT: vaddsubps
+; CHECK: ret
+
+
+define <8 x float> @test6(<8 x float> %A, <8 x float> %B) {
+  %sub = fsub <8 x float> %A, %B
+  %add = fadd <8 x float> %A, %B
+  %vecinit14 = shufflevector <8 x float> %add, <8 x float> %sub, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x float> %vecinit14
+}
+; CHECK-LABEL: test6
+; SSE: xorps
+; SSE-NEXT: addsubps
+; SSE: xorps
+; SSE-NEXT: addsubps
+; AVX: vxorps
+; AVX-NEXT: vaddsubps
+; AVX-NOT: vxorps
+; AVX-NOT: vaddsubps
+; CHECK: ret
+
+
+define <4 x double> @test7(<4 x double> %A, <4 x double> %B) {
+  %sub = fsub <4 x double> %A, %B
+  %add = fadd <4 x double> %A, %B
+  %vecinit6 = shufflevector <4 x double> %add, <4 x double> %sub, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x double> %vecinit6
+}
+; CHECK-LABEL: test7
+; SSE: xorpd
+; SSE-NEXT: addsubpd
+; SSE: xorpd
+; SSE-NEXT: addsubpd
+; AVX: vxorpd
+; AVX-NEXT: vaddsubpd
+; AVX-NOT: vxorpd
+; AVX-NOT: vaddsubpd
+; CHECK: ret
+
+
+define <2 x double> @test8(<2 x double> %A, <2 x double> %B) #0 {
+  %add = fadd <2 x double> %A, %B
+  %sub = fsub <2 x double> %A, %B
+  %vecinit2 = shufflevector <2 x double> %add, <2 x double> %sub, <2 x i32> <i32 0, i32 3>
+  ret <2 x double> %vecinit2
+}
+; CHECK-LABEL: test8
+; SSE: xorpd
+; SSE-NEXT: addsubpd
+; AVX: vxorpd
+; AVX-NEXT: vaddsubpd
+; CHECK: ret
+
+
+define <4 x float> @test5b(<4 x float> %A, <4 x float> %B) {
+  %sub = fsub <4 x float> %A, %B
+  %add = fadd <4 x float> %B, %A
+  %vecinit6 = shufflevector <4 x float> %add, <4 x float> %sub, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %vecinit6
+}
+; CHECK-LABEL: test5
+; SSE: xorps
+; SSE-NEXT: addsubps
+; AVX: vxorps
+; AVX-NEXT: vaddsubps
+; CHECK: ret
+
+
+define <8 x float> @test6b(<8 x float> %A, <8 x float> %B) {
+  %sub = fsub <8 x float> %A, %B
+  %add = fadd <8 x float> %B, %A
+  %vecinit14 = shufflevector <8 x float> %add, <8 x float> %sub, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x float> %vecinit14
+}
+; CHECK-LABEL: test6
+; SSE: xorps
+; SSE-NEXT: addsubps
+; SSE: xorps
+; SSE-NEXT: addsubps
+; AVX: vxorps
+; AVX-NEXT: vaddsubps
+; AVX-NOT: vxorps
+; AVX-NOT: vaddsubps
+; CHECK: ret
+
+
+define <4 x double> @test7b(<4 x double> %A, <4 x double> %B) {
+  %sub = fsub <4 x double> %A, %B
+  %add = fadd <4 x double> %B, %A
+  %vecinit6 = shufflevector <4 x double> %add, <4 x double> %sub, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x double> %vecinit6
+}
+; CHECK-LABEL: test7
+; SSE: xorpd
+; SSE-NEXT: addsubpd
+; SSE: xorpd
+; SSE-NEXT: addsubpd
+; AVX: vxorpd
+; AVX-NEXT: vaddsubpd
+; AVX-NOT: vxorpd
+; AVX-NOT: vaddsubpd
+; CHECK: ret
+
+
+define <2 x double> @test8b(<2 x double> %A, <2 x double> %B) #0 {
+  %add = fadd <2 x double> %B, %A
+  %sub = fsub <2 x double> %A, %B
+  %vecinit2 = shufflevector <2 x double> %add, <2 x double> %sub, <2 x i32> <i32 0, i32 3>
+  ret <2 x double> %vecinit2
+}
+; CHECK-LABEL: test8
+; SSE: xorpd
+; SSE-NEXT: addsubpd
+; AVX: vxorpd
+; AVX-NEXT: vaddsubpd
+; CHECK: ret
+