X86: Custom lower sext v16i8 to v16i16, and the corresponding truncate.

Also update the cost model.

llvm-svn: 193270

8 files changed, 58 insertions, 10 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index d74764714f4..694623e1754 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1150,9 +1150,6 @@ void X86TargetLowering::resetOperationActions() {
     setOperationAction(ISD::FNEG,               MVT::v4f64, Custom);
     setOperationAction(ISD::FABS,               MVT::v4f64, Custom);
 
-    setOperationAction(ISD::TRUNCATE,           MVT::v8i16, Custom);
-    setOperationAction(ISD::TRUNCATE,           MVT::v4i32, Custom);
-
     setOperationAction(ISD::FP_TO_SINT,         MVT::v8i16, Custom);
 
     setOperationAction(ISD::FP_TO_SINT,         MVT::v8i32, Legal);
@@ -1160,8 +1157,6 @@ void X86TargetLowering::resetOperationActions() {
     setOperationAction(ISD::SINT_TO_FP,         MVT::v8i32, Legal);
     setOperationAction(ISD::FP_ROUND,           MVT::v4f32, Legal);
 
-    setOperationAction(ISD::ZERO_EXTEND,        MVT::v16i16, Custom);
-    setOperationAction(ISD::ZERO_EXTEND,        MVT::v8i32, Custom);
     setOperationAction(ISD::UINT_TO_FP,         MVT::v8i8,  Custom);
     setOperationAction(ISD::UINT_TO_FP,         MVT::v8i16, Custom);
 
@@ -1194,10 +1189,16 @@ void X86TargetLowering::resetOperationActions() {
 
     setOperationAction(ISD::SIGN_EXTEND,       MVT::v4i64, Custom);
     setOperationAction(ISD::SIGN_EXTEND,       MVT::v8i32, Custom);
+    setOperationAction(ISD::SIGN_EXTEND,       MVT::v16i16, Custom);
     setOperationAction(ISD::ZERO_EXTEND,       MVT::v4i64, Custom);
     setOperationAction(ISD::ZERO_EXTEND,       MVT::v8i32, Custom);
+    setOperationAction(ISD::ZERO_EXTEND,       MVT::v16i16, Custom);
     setOperationAction(ISD::ANY_EXTEND,        MVT::v4i64, Custom);
     setOperationAction(ISD::ANY_EXTEND,        MVT::v8i32, Custom);
+    setOperationAction(ISD::ANY_EXTEND,        MVT::v16i16, Custom);
+    setOperationAction(ISD::TRUNCATE,          MVT::v16i8, Custom);
+    setOperationAction(ISD::TRUNCATE,          MVT::v8i16, Custom);
+    setOperationAction(ISD::TRUNCATE,          MVT::v4i32, Custom);
 
     if (Subtarget->hasFMA() || Subtarget->hasFMA4()) {
       setOperationAction(ISD::FMA,             MVT::v8f32, Legal);
@@ -10391,7 +10392,8 @@ static SDValue LowerSIGN_EXTEND(SDValue Op, const X86Subtarget *Subtarget,
     return LowerSIGN_EXTEND_AVX512(Op, DAG);
 
   if ((VT != MVT::v4i64 || InVT != MVT::v4i32) &&
-      (VT != MVT::v8i32 || InVT != MVT::v8i16))
+      (VT != MVT::v8i32 || InVT != MVT::v8i16) &&
+      (VT != MVT::v16i16 || InVT != MVT::v16i8))
     return SDValue();
 
   if (Subtarget->hasInt256())
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index 004710b67cc..7cae485b240 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -5602,16 +5602,19 @@ let Predicates = [HasAVX2] in {
 
   def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
   def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
+  def : Pat<(v16i16 (X86vsmovl (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
 }
 
 let Predicates = [HasAVX] in {
   def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
   def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
+  def : Pat<(v8i16 (X86vsmovl (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
 }
 
 let Predicates = [UseSSE41] in {
   def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
   def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
+  def : Pat<(v8i16 (X86vsmovl (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
 }
 
 
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index 36bfeb10aa2..f88a666092b 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -401,12 +401,15 @@ unsigned X86TTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const {
 
   static const TypeConversionCostTblEntry<MVT::SimpleValueType>
   AVXConversionTbl[] = {
+    { ISD::SIGN_EXTEND, MVT::v16i16, MVT::v16i8, 1 },
+    { ISD::ZERO_EXTEND, MVT::v16i16, MVT::v16i8, 1 },
     { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i16, 1 },
     { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i16, 1 },
     { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i32, 1 },
     { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i32, 1 },
     { ISD::TRUNCATE,    MVT::v4i32, MVT::v4i64, 1 },
     { ISD::TRUNCATE,    MVT::v8i16, MVT::v8i32, 1 },
+    { ISD::TRUNCATE,    MVT::v16i8, MVT::v16i16, 2 },
 
     { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i1,  8 },
     { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i8,  8 },
diff --git a/llvm/test/Analysis/CostModel/X86/cast.ll b/llvm/test/Analysis/CostModel/X86/cast.ll
index b69b3bf6304..f3c1283c7e3 100644
--- a/llvm/test/Analysis/CostModel/X86/cast.ll
+++ b/llvm/test/Analysis/CostModel/X86/cast.ll
@@ -38,6 +38,10 @@ define i32 @zext_sext(<8 x i1> %in) {
   ;CHECK: cost of 9 {{.*}} sext
   %S = sext <8 x i1> %in to <8 x i32>
 
+  ;CHECK: cost of 1 {{.*}} zext
+  %A1 = zext <16 x i8> undef to <16 x i16>
+  ;CHECK: cost of 1 {{.*}} sext
+  %A2 = sext <16 x i8> undef to <16 x i16>
   ;CHECK: cost of 1 {{.*}} sext
   %A = sext <8 x i16> undef to <8 x i32>
   ;CHECK: cost of 1 {{.*}} zext
@@ -51,11 +55,13 @@ define i32 @zext_sext(<8 x i1> %in) {
 
   ;CHECK: cost of 1 {{.*}} zext
   %D = zext <4 x i32> undef to <4 x i64>
-  ;CHECK: cost of 1 {{.*}} trunc
 
+  ;CHECK: cost of 1 {{.*}} trunc
   %E = trunc <4 x i64> undef to <4 x i32>
   ;CHECK: cost of 1 {{.*}} trunc
   %F = trunc <8 x i32> undef to <8 x i16>
+  ;CHECK: cost of 2 {{.*}} trunc
+  %F1 = trunc <16 x i16> undef to <16 x i8>
 
   ;CHECK: cost of 3 {{.*}} trunc
   %G = trunc <8 x i64> undef to <8 x i32>
diff --git a/llvm/test/CodeGen/X86/avx-sext.ll b/llvm/test/CodeGen/X86/avx-sext.ll
index b9c70005100..fb2287f5289 100644
--- a/llvm/test/CodeGen/X86/avx-sext.ll
+++ b/llvm/test/CodeGen/X86/avx-sext.ll
@@ -154,6 +154,17 @@ define <4 x i64> @sext_4i1_to_4i64(<4 x i1> %mask) {
   ret <4 x i64> %extmask
 }
 
+; AVX-LABEL: sext_16i8_to_16i16
+; AVX: vpmovsxbw
+; AVX: vmovhlps
+; AVX: vpmovsxbw
+; AVX: ret
+define <16 x i16> @sext_16i8_to_16i16(<16 x i8> *%ptr) {
+ %X = load <16 x i8>* %ptr
+ %Y = sext <16 x i8> %X to <16 x i16>
+ ret <16 x i16> %Y
+}
+
 ; AVX: sext_4i8_to_4i64
 ; AVX: vpslld  $24
 ; AVX: vpsrad  $24
diff --git a/llvm/test/CodeGen/X86/avx-trunc.ll b/llvm/test/CodeGen/X86/avx-trunc.ll
index d0077366444..58d0a356909 100644
--- a/llvm/test/CodeGen/X86/avx-trunc.ll
+++ b/llvm/test/CodeGen/X86/avx-trunc.ll
@@ -12,4 +12,9 @@ define <8 x i16> @trunc_32_16(<8 x i32> %A) nounwind uwtable readnone ssp{
   %B = trunc <8 x i32> %A to <8 x i16>
   ret <8 x i16>%B
 }
-
+define <16 x i8> @trunc_16_8(<16 x i16> %A) nounwind uwtable readnone ssp{
+; CHECK-LABEL: trunc_16_8
+; CHECK: pshufb
+  %B = trunc <16 x i16> %A to <16 x i8>
+  ret <16 x i8> %B
+}
diff --git a/llvm/test/CodeGen/X86/avx2-conversions.ll b/llvm/test/CodeGen/X86/avx2-conversions.ll
index 0143f18fe24..f49718e4c8b 100644
--- a/llvm/test/CodeGen/X86/avx2-conversions.ll
+++ b/llvm/test/CodeGen/X86/avx2-conversions.ll
@@ -72,6 +72,25 @@ define <16 x i16> @zext_16i8_16i16(<16 x i8> %z) {
   ret <16 x i16> %t
 }
 
+; CHECK-LABEL: sext_16i8_16i16:
+; CHECK: vpmovsxbw
+; CHECK-NOT: vinsert
+; CHECK: ret
+define <16 x i16> @sext_16i8_16i16(<16 x i8> %z) {
+  %t = sext <16 x i8> %z to <16 x i16>
+  ret <16 x i16> %t
+}
+
+; CHECK-LABEL: trunc_16i16_16i8:
+; CHECK: vpshufb
+; CHECK: vpshufb
+; CHECK: vpor
+; CHECK: ret
+define <16 x i8> @trunc_16i16_16i8(<16 x i16> %z) {
+  %t = trunc <16 x i16> %z to <16 x i8>
+  ret <16 x i8> %t
+}
+
 ; CHECK: load_sext_test1
 ; CHECK: vpmovsxdq (%r{{[^,]*}}), %ymm{{.*}}
 ; CHECK: ret 
diff --git a/llvm/test/CodeGen/X86/pmovsx-inreg.ll b/llvm/test/CodeGen/X86/pmovsx-inreg.ll
index d30d7d0694a..07979f61ddd 100644
--- a/llvm/test/CodeGen/X86/pmovsx-inreg.ll
+++ b/llvm/test/CodeGen/X86/pmovsx-inreg.ll
@@ -86,8 +86,7 @@ define void @test6(<16 x i8>* %in, <16 x i16>* %out) nounwind {
   ret void
 
 ; AVX2-LABEL: test6:
-; FIXME: v16i8 -> v16i16 is scalarized.
-; AVX2-NOT: pmovsx
+; AVX2: vpmovsxbw
 }
 
 define void @test7(<2 x i16>* %in, <2 x i64>* %out) nounwind {