[x86] Add store-folded memop patterns for vcvtps2ph

Differential Revision: http://reviews.llvm.org/D7296

llvm-svn: 235517

diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index 6366b1a..dfaf9c2 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -8243,6 +8243,18 @@ let Predicates = [HasF16C] in {
             (VCVTPH2PSrm addr:$src)>;
   def : Pat<(int_x86_vcvtph2ps_128 (vzload_v2i64 addr:$src)),
             (VCVTPH2PSrm addr:$src)>;
+
+  def : Pat<(store (f64 (vector_extract (bc_v2f64 (v8i16
+                  (int_x86_vcvtps2ph_128 VR128:$src1, i32:$src2))), (iPTR 0))),
+                   addr:$dst),
+                   (VCVTPS2PHmr addr:$dst, VR128:$src1, imm:$src2)>;
+  def : Pat<(store (i64 (vector_extract (bc_v2i64 (v8i16
+                  (int_x86_vcvtps2ph_128 VR128:$src1, i32:$src2))), (iPTR 0))),
+                   addr:$dst),
+                   (VCVTPS2PHmr addr:$dst, VR128:$src1, imm:$src2)>;
+  def : Pat<(store (v8i16 (int_x86_vcvtps2ph_256 VR256:$src1, i32:$src2)),
+                   addr:$dst),
+                   (VCVTPS2PHYmr addr:$dst, VR256:$src1, imm:$src2)>;
 }
 
 // Patterns for  matching conversions from float to half-float and vice versa.

diff --git a/llvm/test/CodeGen/X86/f16c-intrinsics.ll b/llvm/test/CodeGen/X86/f16c-intrinsics.ll
index 02967d5..485592a 100644 (file)
--- a/llvm/test/CodeGen/X86/f16c-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/f16c-intrinsics.ll
@@ -1,8 +1,8 @@
-; RUN: llc < %s -march=x86 -mattr=+avx,+f16c | FileCheck %s
-; RUN: llc < %s -march=x86-64 -mattr=+avx,+f16c | FileCheck %s
+; RUN: llc < %s -mtriple=i686-unknown-unknown   -mattr=+avx,+f16c | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+f16c | FileCheck %s
 
 define <4 x float> @test_x86_vcvtph2ps_128(<8 x i16> %a0) {
-  ; CHECK-LABEL: test_x86_vcvtph2ps_128
+  ; CHECK-LABEL: test_x86_vcvtph2ps_128:
   ; CHECK-NOT: vmov
   ; CHECK: vcvtph2ps
   %res = call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> %a0) ; <<4 x float>> [#uses=1]
@@ -12,7 +12,7 @@ declare <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16>) nounwind readonly
 
 
 define <8 x float> @test_x86_vcvtph2ps_256(<8 x i16> %a0) {
-  ; CHECK-LABEL: test_x86_vcvtph2ps_256
+  ; CHECK-LABEL: test_x86_vcvtph2ps_256:
   ; CHECK-NOT: vmov
   ; CHECK: vcvtph2ps
   %res = call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> %a0) ; <<8 x float>> [#uses=1]
@@ -31,7 +31,7 @@ entry:
 }
 
 define <8 x i16> @test_x86_vcvtps2ph_128(<4 x float> %a0) {
-  ; CHECK-LABEL: test_x86_vcvtps2ph_128
+  ; CHECK-LABEL: test_x86_vcvtps2ph_128:
   ; CHECK-NOT: vmov
   ; CHECK: vcvtps2ph
   %res = call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %a0, i32 0) ; <<8 x i16>> [#uses=1]
@@ -39,9 +39,8 @@ define <8 x i16> @test_x86_vcvtps2ph_128(<4 x float> %a0) {
 }
 declare <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float>, i32) nounwind readonly
 
-
 define <8 x i16> @test_x86_vcvtps2ph_256(<8 x float> %a0) {
-  ; CHECK-LABEL: test_x86_vcvtps2ph_256
+  ; CHECK-LABEL: test_x86_vcvtps2ph_256:
   ; CHECK-NOT: vmov
   ; CHECK: vcvtps2ph
   %res = call <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float> %a0, i32 0) ; <<8 x i16>> [#uses=1]
@@ -50,7 +49,7 @@ define <8 x i16> @test_x86_vcvtps2ph_256(<8 x float> %a0) {
 declare <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float>, i32) nounwind readonly
 
 define <4 x float> @test_x86_vcvtps2ph_128_scalar(i64* %ptr) {
-; CHECK-LABEL: test_x86_vcvtps2ph_128_scalar
+; CHECK-LABEL: test_x86_vcvtps2ph_128_scalar:
 ; CHECK-NOT: vmov
 ; CHECK: vcvtph2ps (%
 
@@ -61,3 +60,48 @@ define <4 x float> @test_x86_vcvtps2ph_128_scalar(i64* %ptr) {
   %res = tail call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> %bc) #2
   ret <4 x float> %res
 }
+
+define void @test_x86_vcvtps2ph_256_m(<8 x i16>* nocapture %d, <8 x float> %a) nounwind {
+entry:
+  ; CHECK-LABEL: test_x86_vcvtps2ph_256_m:
+  ; CHECK-NOT: vmov
+  ; CHECK: vcvtps2ph  $3, %ymm0, (%
+  %0 = tail call <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float> %a, i32 3)
+  store <8 x i16> %0, <8 x i16>* %d, align 16
+  ret void
+}
+
+define void @test_x86_vcvtps2ph_128_m(<4 x i16>* nocapture %d, <4 x float> %a) nounwind {
+entry:
+  ; CHECK-LABEL: test_x86_vcvtps2ph_128_m:
+  ; CHECK-NOT: vmov
+  ; CHECK: vcvtps2ph  $3, %xmm0, (%
+  %0 = tail call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %a, i32 3)
+  %1 = shufflevector <8 x i16> %0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  store <4 x i16> %1, <4 x i16>* %d, align 8
+  ret void
+}
+
+define void @test_x86_vcvtps2ph_128_m2(double* nocapture %hf4x16, <4 x float> %f4x32) #0 {
+entry:
+  ; CHECK-LABEL: test_x86_vcvtps2ph_128_m2:
+  ; CHECK-NOT: vmov
+  ; CHECK: vcvtps2ph  $3, %xmm0, (%
+  %0 = tail call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %f4x32, i32 3)
+  %1 = bitcast <8 x i16> %0 to <2 x double>
+  %vecext = extractelement <2 x double> %1, i32 0
+  store double %vecext, double* %hf4x16, align 8
+  ret void
+}
+
+define void @test_x86_vcvtps2ph_128_m3(i64* nocapture %hf4x16, <4 x float> %f4x32) #0 {
+entry:
+  ; CHECK-LABEL: test_x86_vcvtps2ph_128_m3:
+  ; CHECK-NOT: vmov
+  ; CHECK: vcvtps2ph  $3, %xmm0, (%
+  %0 = tail call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %f4x32, i32 3)
+  %1 = bitcast <8 x i16> %0 to <2 x i64>
+  %vecext = extractelement <2 x i64> %1, i32 0
+  store i64 %vecext, i64* %hf4x16, align 8
+  ret void
+}