Add AVX versions of blend vector operations and fix some issues noticed

in Nadav's r139285 and r139287 commits.

1) Rename vsel.ll to a more descriptive name
2) Change the order of BLEND operands to "Op1, Op2, Cond", this is
necessary because PBLENDVB is already used in different places with
this order, and it was being emitted in the wrong way for vselect
3) Add AVX patterns and tests for the same SSE41 instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139305 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 69 insertions, 15 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index d74a872..cf90490 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8697,7 +8697,7 @@ SDValue X86TargetLowering::LowerVSELECT(SDValue Op, SelectionDAG &DAG) const {
   SDValue Op2 = Op.getOperand(2);
   DebugLoc DL = Op.getDebugLoc();
 
-  SDValue Ops[] = {Cond, Op1, Op2};
+  SDValue Ops[] = {Op1, Op2, Cond};
 
   assert(Op1.getValueType().isVector() && "Op1 must be a vector");
   assert(Op2.getValueType().isVector() && "Op2 must be a vector");
diff --git a/lib/Target/X86/X86InstrFragmentsSIMD.td b/lib/Target/X86/X86InstrFragmentsSIMD.td
index 7ad9c87..c2db917 100644
--- a/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -61,10 +61,10 @@ def X86psignd  : SDNode<"X86ISD::PSIGND",
 def X86pblendvb : SDNode<"X86ISD::PBLENDVB",
                  SDTypeProfile<1, 3, [SDTCisVT<0, v16i8>, SDTCisSameAs<0,1>,
                                       SDTCisSameAs<0,2>, SDTCisSameAs<0,3>]>>;
-def X86blendvpd : SDNode<"X86ISD::BLENDVPD", 
+def X86blendvpd : SDNode<"X86ISD::BLENDVPD",
                   SDTypeProfile<1, 3, [SDTCisVT<0, v2i64>, SDTCisSameAs<0,1>,
                                        SDTCisSameAs<0,2>, SDTCisSameAs<0,3>]>>;
-def X86blendvps : SDNode<"X86ISD::BLENDVPS", 
+def X86blendvps : SDNode<"X86ISD::BLENDVPS",
                  SDTypeProfile<1, 3, [SDTCisVT<0, v4i32>, SDTCisSameAs<0,1>,
                                       SDTCisSameAs<0,2>, SDTCisSameAs<0,3>]>>;
 def X86pextrb  : SDNode<"X86ISD::PEXTRB",
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 9148c76..6bcba72 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -5853,9 +5853,14 @@ defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
                                          memopv32i8, int_x86_avx_blendv_ps_256>;
 
-def : Pat<(X86pblendvb VR128:$src1, VR128:$src2, VR128:$src3),
-          (VPBLENDVBrr VR128:$src1, VR128:$src2, VR128:$src3)>,
-          Requires<[HasAVX]>;
+let Predicates = [HasAVX] in {
+  def : Pat<(X86pblendvb VR128:$src1, VR128:$src2, VR128:$mask),
+            (VPBLENDVBrr VR128:$src1, VR128:$src2, VR128:$mask)>;
+  def : Pat<(X86blendvpd VR128:$src1, VR128:$src2, VR128:$mask),
+            (VBLENDVPDrr VR128:$src1, VR128:$src2, VR128:$mask)>;
+  def : Pat<(X86blendvps VR128:$src1, VR128:$src2, VR128:$mask),
+            (VBLENDVPSrr VR128:$src1, VR128:$src2, VR128:$mask)>;
+}
 
 /// SS41I_ternary_int - SSE 4.1 ternary operator
 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
@@ -5877,16 +5882,18 @@ let Uses = [XMM0], Constraints = "$src1 = $dst" in {
   }
 }
 
-defm BLENDVPD     : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
-defm BLENDVPS     : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
-defm PBLENDVB     : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
+defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
+defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
+defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
 
-def : Pat<(X86pblendvb VR128:$src1, VR128:$src2, XMM0),
-          (PBLENDVBrr0 VR128:$src1, VR128:$src2)>, Requires<[HasSSE41]>;
-def : Pat<(X86blendvpd  XMM0, VR128:$src1, VR128:$src2),
-          (BLENDVPDrr0 VR128:$src1, VR128:$src2)>, Requires<[HasSSE41]>;
-def : Pat<(X86blendvps  XMM0, VR128:$src1, VR128:$src2),
-          (BLENDVPSrr0 VR128:$src1, VR128:$src2)>, Requires<[HasSSE41]>;
+let Predicates = [HasSSE41] in {
+  def : Pat<(X86pblendvb VR128:$src1, VR128:$src2, XMM0),
+            (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
+  def : Pat<(X86blendvpd VR128:$src1, VR128:$src2, XMM0),
+            (BLENDVPDrr0 VR128:$src1, VR128:$src2)>;
+  def : Pat<(X86blendvps VR128:$src1, VR128:$src2, XMM0),
+            (BLENDVPSrr0 VR128:$src1, VR128:$src2)>;
+}
 
 let Predicates = [HasAVX] in
 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
diff --git a/test/CodeGen/X86/avx-blend.ll b/test/CodeGen/X86/avx-blend.ll
new file mode 100644
index 0000000..68289ad
--- /dev/null
+++ b/test/CodeGen/X86/avx-blend.ll
@@ -0,0 +1,47 @@
+; RUN: llc < %s -mattr=+avx -march=x86 | FileCheck %s
+
+;CHECK: vsel_float
+;CHECK: vblendvps
+;CHECK: ret
+define <4 x float> @vsel_float(<4 x float> %v1, <4 x float> %v2) {
+  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x float> %v1, <4 x float> %v2
+  ret <4 x float> %vsel
+}
+
+
+;CHECK: vsel_i32
+;CHECK: vblendvps
+;CHECK: ret
+define <4 x i32> @vsel_i32(<4 x i32> %v1, <4 x i32> %v2) {
+  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x i32> %v1, <4 x i32> %v2
+  ret <4 x i32> %vsel
+}
+
+
+;CHECK: vsel_double
+;CHECK: vblendvpd
+;CHECK: ret
+define <2 x double> @vsel_double(<2 x double> %v1, <2 x double> %v2) {
+  %vsel = select <2 x i1> <i1 true, i1 false>, <2 x double> %v1, <2 x double> %v2
+  ret <2 x double> %vsel
+}
+
+
+;CHECK: vsel_i64
+;CHECK: vblendvpd
+;CHECK: ret
+define <2 x i64> @vsel_i64(<2 x i64> %v1, <2 x i64> %v2) {
+  %vsel = select <2 x i1> <i1 true, i1 false>, <2 x i64> %v1, <2 x i64> %v2
+  ret <2 x i64> %vsel
+}
+
+
+;CHECK: vsel_i8
+;CHECK: vpblendvb
+;CHECK: ret
+define <16 x i8> @vsel_i8(<16 x i8> %v1, <16 x i8> %v2) {
+  %vsel = select <16 x i1> <i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 false, i1 false>, <16 x i8> %v1, <16 x i8> %v2
+  ret <16 x i8> %vsel
+}
+
+
diff --git a/test/CodeGen/X86/vsel.ll b/test/CodeGen/X86/sse41-blend.ll
index 3c854ac..3c854ac 100644
--- a/test/CodeGen/X86/vsel.ll
+++ b/test/CodeGen/X86/sse41-blend.ll