Lower the fcmp instruction for <4 x float> operands.

crosstest/test_fcmp.pnacl.ll

Index: crosstest/test_fcmp.pnacl.ll
diff --git a/crosstest/test_fcmp.pnacl.ll b/crosstest/test_fcmp.pnacl.ll
index 7c4d42e5a15f99bd80ca642b428907088704c4cc..a175b1deb4741825541565c13eb6f70c1c603a56 100644
--- a/crosstest/test_fcmp.pnacl.ll
+++ b/crosstest/test_fcmp.pnacl.ll
@@ -1,7 +1,7 @@
 target triple = "i686-pc-linux-gnu"
 
-; This file is extracted from fp.pnacl.ll in the lit tests, with
-; the "internal" attribute removed from the functions.
+; This file is extracted from fp.pnacl.ll and vector-fcmp.ll in the lit
+; tests, with the "internal" attribute removed from the functions.
 
 define i32 @fcmpFalseFloat(float %a, float %b) {
 entry:
@@ -322,3 +322,151 @@ entry:
 }
 ; CHECK: fcmpTrueDouble:
 ; CHECK: mov {{.*}}, 1
+
+define <4 x i32> @fcmpFalseVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp false <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpFalseVector:
+; CHECK: pxor
+}
+
+define <4 x i32> @fcmpOeqVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp oeq <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpOeqVector:
+; CHECK: cmpeqps
+}
+
+define <4 x i32> @fcmpOgeVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp oge <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpOgeVector:
+; CHECK: cmpleps
+}
+
+define <4 x i32> @fcmpOgtVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp ogt <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpOgtVector:
+; CHECK: cmpltps
+}
+
+define <4 x i32> @fcmpOleVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp ole <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpOleVector:
+; CHECK: cmpleps
+}
+
+define <4 x i32> @fcmpOltVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp olt <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpOltVector:
+; CHECK: cmpltps
+}
+
+define <4 x i32> @fcmpOneVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp one <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpOneVector:
+; CHECK: cmpneqps
+; CHECK: cmpordps
+; CHECK: pand
+}
+
+define <4 x i32> @fcmpOrdVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp ord <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpOrdVector:
+; CHECK: cmpordps
+}
+
+define <4 x i32> @fcmpTrueVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp true <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpTrueVector:
+; CHECK: pcmpeqd
+}
+
+define <4 x i32> @fcmpUeqVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp ueq <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpUeqVector:
+; CHECK: cmpeqps
+; CHECK: cmpunordps
+; CHECK: por
+}
+
+define <4 x i32> @fcmpUgeVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp uge <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpUgeVector:
+; CHECK: cmpnltps
+}
+
+define <4 x i32> @fcmpUgtVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp ugt <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpUgtVector:
+; CHECK: cmpnleps
+}
+
+define <4 x i32> @fcmpUleVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp ule <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpUleVector:
+; CHECK: cmpnltps
+}
+
+define <4 x i32> @fcmpUltVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp ult <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpUltVector:
+; CHECK: cmpnleps
+}
+
+define <4 x i32> @fcmpUneVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp une <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpUneVector:
+; CHECK: cmpneqps
+}
+
+define <4 x i32> @fcmpUnoVector(<4 x float> %a, <4 x float> %b) {
+entry:
+  %res.trunc = fcmp uno <4 x float> %a, %b
+  %res = sext <4 x i1> %res.trunc to <4 x i32>
+  ret <4 x i32> %res
+; CHECK-LABEL: fcmpUnoVector:
+; CHECK: cmpunordps
+}