[SelectionDAG][X86] Teach promotion legalization for fp_to_sint/fp_to_uint to insert an assertsext/assertzext based on the original type

If we put in an assertsext/zext here, we're able to generate better truncate code using pack on pre-avx512 targets.

Similar is already done during type legalization. This is the equivalent for op legalization

Differential Revision: https://reviews.llvm.org/D40591

llvm-svn: 319368

3 files changed, 16 insertions, 23 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
index 928db8280bb..74970ab5792 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
@@ -512,9 +512,17 @@ SDValue VectorLegalizer::PromoteFP_TO_INT(SDValue Op, bool isSigned) {
     }
   }
 
-  SDLoc loc(Op);
-  SDValue promoted  = DAG.getNode(NewOpc, SDLoc(Op), NewVT, Op.getOperand(0));
-  return DAG.getNode(ISD::TRUNCATE, SDLoc(Op), VT, promoted);
+  SDLoc dl(Op);
+  SDValue Promoted  = DAG.getNode(NewOpc, dl, NewVT, Op.getOperand(0));
+
+  // Assert that the converted value fits in the original type.  If it doesn't
+  // (eg: because the value being converted is too big), then the result of the
+  // original operation was undefined anyway, so the assert is still correct.
+  Promoted = DAG.getNode(Op->getOpcode() == ISD::FP_TO_UINT ? ISD::AssertZext
+                                                            : ISD::AssertSext,
+                         dl, NewVT, Promoted,
+                         DAG.getValueType(VT.getScalarType()));
+  return DAG.getNode(ISD::TRUNCATE, dl, VT, Promoted);
 }
 
 SDValue VectorLegalizer::ExpandLoad(SDValue Op) {
diff --git a/llvm/test/CodeGen/X86/avx-cvt-2.ll b/llvm/test/CodeGen/X86/avx-cvt-2.ll
index c955756811f..f38127960bf 100644
--- a/llvm/test/CodeGen/X86/avx-cvt-2.ll
+++ b/llvm/test/CodeGen/X86/avx-cvt-2.ll
@@ -12,10 +12,7 @@ define void @fptoui16(%f32vec_t %a, %i16vec_t *%p) {
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    vmovdqa %xmm0, (%rdi)
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
@@ -29,10 +26,7 @@ define void @fptosi16(%f32vec_t %a, %i16vec_t *%p) {
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    vmovdqa %xmm0, (%rdi)
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
@@ -46,10 +40,7 @@ define void @fptoui8(%f32vec_t %a, %i8vec_t *%p) {
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; CHECK-NEXT:    vmovq %xmm0, (%rdi)
 ; CHECK-NEXT:    vzeroupper
@@ -64,10 +55,7 @@ define void @fptosi8(%f32vec_t %a, %i8vec_t *%p) {
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; CHECK-NEXT:    vmovq %xmm0, (%rdi)
 ; CHECK-NEXT:    vzeroupper
diff --git a/llvm/test/CodeGen/X86/vec_cast2.ll b/llvm/test/CodeGen/X86/vec_cast2.ll
index c03b330b88e..723b0ca53fc 100644
--- a/llvm/test/CodeGen/X86/vec_cast2.ll
+++ b/llvm/test/CodeGen/X86/vec_cast2.ll
@@ -88,10 +88,7 @@ define <8 x i8> @foo3_8(<8 x float> %src) {
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vcvttps2dq %ymm0, %ymm0
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; CHECK-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; CHECK-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; CHECK-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retl
 ;