[X86][SSE] Always combine target shuffles to MOVSD/MOVSS

Now we can commute to BLENDPD/BLENDPS on SSE41+ targets if necessary, so simplify the combine matching where we can.

This required me to add a couple of scalar math movsd/moss fold patterns that hadn't been needed in the past.

llvm-svn: 283038

6 files changed, 73 insertions, 31 deletions

diff --git a/llvm/test/CodeGen/X86/sse-scalar-fp-arith.ll b/llvm/test/CodeGen/X86/sse-scalar-fp-arith.ll
index 1fe5b54d2bb..228e7827537 100644
--- a/llvm/test/CodeGen/X86/sse-scalar-fp-arith.ll
+++ b/llvm/test/CodeGen/X86/sse-scalar-fp-arith.ll
@@ -91,11 +91,17 @@ define <4 x float> @test_sqrt_ss(<4 x float> %a) {
 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: test_sqrt_ss:
-; AVX:       # BB#0:
-; AVX-NEXT:    vsqrtss %xmm0, %xmm0, %xmm1
-; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
-; AVX-NEXT:    retq
+; AVX1-LABEL: test_sqrt_ss:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vsqrtss %xmm0, %xmm0, %xmm1
+; AVX1-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
+; AVX1-NEXT:    retq
+;
+; AVX512-LABEL: test_sqrt_ss:
+; AVX512:       # BB#0:
+; AVX512-NEXT:    vsqrtss %xmm0, %xmm0, %xmm1
+; AVX512-NEXT:    vmovss {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
+; AVX512-NEXT:    retq
   %1 = extractelement <4 x float> %a, i32 0
   %2 = call float @llvm.sqrt.f32(float %1)
   %3 = insertelement <4 x float> %a, float %2, i32 0
diff --git a/llvm/test/CodeGen/X86/sse41.ll b/llvm/test/CodeGen/X86/sse41.ll
index 3cb754c8f93..4a009023a7b 100644
--- a/llvm/test/CodeGen/X86/sse41.ll
+++ b/llvm/test/CodeGen/X86/sse41.ll
@@ -177,7 +177,8 @@ define <4 x float> @blendps_not_insertps_1(<4 x float> %t1, float %t2) nounwind
 define <4 x float> @insertps_or_blendps(<4 x float> %t1, float %t2) minsize nounwind {
 ; X32-LABEL: insertps_or_blendps:
 ; X32:       ## BB#0:
-; X32-NEXT:    insertps {{.*#+}} xmm0 = mem[0],xmm0[1,2,3]
+; X32-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: insertps_or_blendps:
diff --git a/llvm/test/CodeGen/X86/vector-blend.ll b/llvm/test/CodeGen/X86/vector-blend.ll
index 309fa98145c..e07d7c3e8c7 100644
--- a/llvm/test/CodeGen/X86/vector-blend.ll
+++ b/llvm/test/CodeGen/X86/vector-blend.ll
@@ -729,8 +729,8 @@ define <8 x float> @blend_shufflevector_8xfloat(<8 x float> %a, <8 x float> %b)
 ;
 ; SSE41-LABEL: blend_shufflevector_8xfloat:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3]
 ; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm3[0,1],xmm1[2],xmm3[3]
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: blend_shufflevector_8xfloat:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-128-v2.ll b/llvm/test/CodeGen/X86/vector-shuffle-128-v2.ll
index 6ef383c858e..71bee75507c 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-128-v2.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-128-v2.ll
@@ -246,10 +246,20 @@ define <2 x double> @shuffle_v2f64_03(<2 x double> %a, <2 x double> %b) {
 ; SSE41-NEXT:    blendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v2f64_03:
-; AVX:       # BB#0:
-; AVX-NEXT:    vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v2f64_03:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v2f64_03:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v2f64_03:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 0, i32 3>
   ret <2 x double> %shuffle
 }
@@ -274,10 +284,20 @@ define <2 x double> @shuffle_v2f64_21(<2 x double> %a, <2 x double> %b) {
 ; SSE41-NEXT:    blendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v2f64_21:
-; AVX:       # BB#0:
-; AVX-NEXT:    vblendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v2f64_21:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v2f64_21:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vblendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v2f64_21:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    vmovsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 2, i32 1>
   ret <2 x double> %shuffle
 }
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
index c286fd426b1..16b2efe88a4 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
@@ -49,16 +49,23 @@ define <2 x double> @combine_pshufb_as_movsd(<2 x double> %a0, <2 x double> %a1)
 ;
 ; SSE41-LABEL: combine_pshufb_as_movsd:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
-; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[8,9,10,11,12,13,14,15,0,1,2,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    blendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: combine_pshufb_as_movsd:
-; AVX:       # BB#0:
-; AVX-NEXT:    vshufpd {{.*#+}} xmm0 = xmm1[1],xmm0[0]
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX-NEXT:    retq
+; AVX1-LABEL: combine_pshufb_as_movsd:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_pshufb_as_movsd:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; AVX2-NEXT:    retq
+;
+; AVX512F-LABEL: combine_pshufb_as_movsd:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; AVX512F-NEXT:    retq
   %1 = shufflevector <2 x double> %a0, <2 x double> %a1, <2 x i32> <i32 3, i32 0>
   %2 = bitcast <2 x double> %1 to <16 x i8>
   %3 = tail call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %2, <16 x i8> <i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7>)
@@ -77,10 +84,20 @@ define <4 x float> @combine_pshufb_as_movss(<4 x float> %a0, <4 x float> %a1) {
 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: combine_pshufb_as_movss:
-; AVX:       # BB#0:
-; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
-; AVX-NEXT:    retq
+; AVX1-LABEL: combine_pshufb_as_movss:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_pshufb_as_movss:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
+; AVX2-NEXT:    retq
+;
+; AVX512F-LABEL: combine_pshufb_as_movss:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vmovss {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
+; AVX512F-NEXT:    retq
   %1 = shufflevector <4 x float> %a0, <4 x float> %a1, <4 x i32> <i32 4, i32 3, i32 2, i32 1>
   %2 = bitcast <4 x float> %1 to <16 x i8>
   %3 = tail call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %2, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 12, i8 13, i8 14, i8 15, i8 8, i8 9, i8 10, i8 11, i8 4, i8 5, i8 6, i8 7>)
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-xop.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-xop.ll
index 121fafc84c5..c1d4446023d 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-xop.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-xop.ll
@@ -262,15 +262,13 @@ define <2 x double> @constant_fold_vpermil2pd() {
 ; X32-LABEL: constant_fold_vpermil2pd:
 ; X32:       # BB#0:
 ; X32-NEXT:    vmovapd {{.*#+}} xmm0 = [-2.000000e+00,-1.000000e+00]
-; X32-NEXT:    vmovapd {{.*#+}} xmm1 = [1.000000e+00,2.000000e+00]
-; X32-NEXT:    vpermil2pd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; X32-NEXT:    vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1]
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: constant_fold_vpermil2pd:
 ; X64:       # BB#0:
 ; X64-NEXT:    vmovapd {{.*#+}} xmm0 = [-2.000000e+00,-1.000000e+00]
-; X64-NEXT:    vmovapd {{.*#+}} xmm1 = [1.000000e+00,2.000000e+00]
-; X64-NEXT:    vpermil2pd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; X64-NEXT:    vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1]
 ; X64-NEXT:    retq
   %1 = call <2 x double> @llvm.x86.xop.vpermil2pd(<2 x double> <double 1.0, double 2.0>, <2 x double> <double -2.0, double -1.0>, <2 x i64> <i64 4, i64 2>, i8 2)
   ret <2 x double> %1