[Power9] Optimize codgen for conversions of int to float128

Optimize code sequences for integer conversion to fp128 when the integer is a result of:
  * float->int
  * float->long
  * double->int
  * double->long

Differential Revision: https://reviews.llvm.org/D48429

llvm-svn: 336316

2 files changed, 136 insertions, 10 deletions

diff --git a/llvm/test/CodeGen/PowerPC/f128-conv.ll b/llvm/test/CodeGen/PowerPC/f128-conv.ll
index d330bd777bd..42949685e3c 100644
--- a/llvm/test/CodeGen/PowerPC/f128-conv.ll
+++ b/llvm/test/CodeGen/PowerPC/f128-conv.ll
@@ -714,3 +714,129 @@ entry:
   store fp128 %conv, fp128* %res, align 16
   ret void
 }
+
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+
+; Function Attrs: norecurse nounwind
+define void @cvdp2sw2qp(double %val, fp128* nocapture %res) {
+; CHECK-LABEL: cvdp2sw2qp:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xscvdpsxws 2, 1
+; CHECK-NEXT:    vextsw2d 2, 2
+; CHECK-NEXT:    xscvsdqp 2, 2
+; CHECK-NEXT:    stxv 2, 0(4)
+; CHECK-NEXT:    blr
+entry:
+  %conv = fptosi double %val to i32
+  %conv1 = sitofp i32 %conv to fp128
+  store fp128 %conv1, fp128* %res, align 16
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define void @cvdp2sdw2qp(double %val, fp128* nocapture %res) {
+; CHECK-LABEL: cvdp2sdw2qp:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xscvdpsxds 2, 1
+; CHECK-NEXT:    xscvsdqp 2, 2
+; CHECK-NEXT:    stxv 2, 0(4)
+; CHECK-NEXT:    blr
+entry:
+  %conv = fptosi double %val to i64
+  %conv1 = sitofp i64 %conv to fp128
+  store fp128 %conv1, fp128* %res, align 16
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define void @cvsp2sw2qp(float %val, fp128* nocapture %res) {
+; CHECK-LABEL: cvsp2sw2qp:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xscvdpsxws 2, 1
+; CHECK-NEXT:    vextsw2d 2, 2
+; CHECK-NEXT:    xscvsdqp 2, 2
+; CHECK-NEXT:    stxv 2, 0(4)
+; CHECK-NEXT:    blr
+entry:
+  %conv = fptosi float %val to i32
+  %conv1 = sitofp i32 %conv to fp128
+  store fp128 %conv1, fp128* %res, align 16
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define void @cvsp2sdw2qp(float %val, fp128* nocapture %res) {
+; CHECK-LABEL: cvsp2sdw2qp:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xscvdpsxds 2, 1
+; CHECK-NEXT:    xscvsdqp 2, 2
+; CHECK-NEXT:    stxv 2, 0(4)
+; CHECK-NEXT:    blr
+entry:
+  %conv = fptosi float %val to i64
+  %conv1 = sitofp i64 %conv to fp128
+  store fp128 %conv1, fp128* %res, align 16
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define void @cvdp2uw2qp(double %val, fp128* nocapture %res) {
+; CHECK-LABEL: cvdp2uw2qp:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xscvdpuxws 0, 1
+; CHECK-NEXT:    xxextractuw 2, 0, 8
+; CHECK-NEXT:    xscvudqp 2, 2
+; CHECK-NEXT:    stxv 2, 0(4)
+; CHECK-NEXT:    blr
+entry:
+  %conv = fptoui double %val to i32
+  %conv1 = uitofp i32 %conv to fp128
+  store fp128 %conv1, fp128* %res, align 16
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define void @cvdp2udw2qp(double %val, fp128* nocapture %res) {
+; CHECK-LABEL: cvdp2udw2qp:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xscvdpuxds 2, 1
+; CHECK-NEXT:    xscvudqp 2, 2
+; CHECK-NEXT:    stxv 2, 0(4)
+; CHECK-NEXT:    blr
+entry:
+  %conv = fptoui double %val to i64
+  %conv1 = uitofp i64 %conv to fp128
+  store fp128 %conv1, fp128* %res, align 16
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define void @cvsp2uw2qp(float %val, fp128* nocapture %res) {
+; CHECK-LABEL: cvsp2uw2qp:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xscvdpuxws 0, 1
+; CHECK-NEXT:    xxextractuw 2, 0, 8
+; CHECK-NEXT:    xscvudqp 2, 2
+; CHECK-NEXT:    stxv 2, 0(4)
+; CHECK-NEXT:    blr
+entry:
+  %conv = fptoui float %val to i32
+  %conv1 = uitofp i32 %conv to fp128
+  store fp128 %conv1, fp128* %res, align 16
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define void @cvsp2udw2qp(float %val, fp128* nocapture %res) {
+; CHECK-LABEL: cvsp2udw2qp:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xscvdpuxds 2, 1
+; CHECK-NEXT:    xscvudqp 2, 2
+; CHECK-NEXT:    stxv 2, 0(4)
+; CHECK-NEXT:    blr
+entry:
+  %conv = fptoui float %val to i64
+  %conv1 = uitofp i64 %conv to fp128
+  store fp128 %conv1, fp128* %res, align 16
+  ret void
+}
diff --git a/llvm/test/CodeGen/PowerPC/f128-passByValue.ll b/llvm/test/CodeGen/PowerPC/f128-passByValue.ll
index a495b5b48a9..ed5edc1acd8 100644
--- a/llvm/test/CodeGen/PowerPC/f128-passByValue.ll
+++ b/llvm/test/CodeGen/PowerPC/f128-passByValue.ll
@@ -218,12 +218,12 @@ entry:
 define void @mixParam_03(fp128 %f1, double* nocapture %d1, <4 x i32> %vec1,
 ; CHECK-LABEL: mixParam_03:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    ld 3, 104(1)
-; CHECK-NEXT:    mtvsrwa [[REG2:[0-9]+]], 10
-; CHECK-NEXT:    stxv [[REG0:[0-9]+]], 0(9)
-; CHECK-NEXT:    stxvx [[REG1:[0-9]+]], 0, 3
-; CHECK-NEXT:    lxv [[REG0]], 0(9)
-; CHECK-NEXT:    xscvsdqp [[REG1]], [[REG2]]
+; CHECK-DAG:     ld 3, 104(1)
+; CHECK-DAG:     mtvsrwa [[REG2:[0-9]+]], 10
+; CHECK-DAG:     stxv 2, 0(9)
+; CHECK-DAG:     xscvsdqp [[REG1:[0-9]+]], [[REG2]]
+; CHECK:         stxvx 3, 0, 3
+; CHECK-NEXT:    lxv 2, 0(9)
 ; CHECK-NEXT:    xsaddqp 2, 2, [[REG1]]
 ; CHECK-NEXT:    xscvqpdp 2, 2
 ; CHECK-NEXT:    stxsd 2, 0(5)
@@ -249,10 +249,10 @@ define fastcc void @mixParam_03f(fp128 %f1, double* nocapture %d1, <4 x i32> %ve
 ; CHECK-NEXT:    stxv [[REG1:[0-9]+]], 0(4)
 ; CHECK-NEXT:    stxv [[REG2:[0-9]+]], 0(7)
 ; CHECK-NEXT:    lxv [[REG1]], 0(4)
-; CHECK-NEXT:    xscvsdqp [[REG2]], [[REG0]]
-; CHECK-NEXT:    xsaddqp [[REG1]], [[REG1]], [[REG2]]
-; CHECK-NEXT:    xscvqpdp [[REG1]], [[REG1]]
-; CHECK-NEXT:    stxsd [[REG1]], 0(3)
+; CHECK-NEXT:    xscvsdqp [[REG3:[0-9]+]], [[REG0]]
+; CHECK-NEXT:    xsaddqp [[REG4:[0-9]+]], [[REG1]], [[REG3]]
+; CHECK-NEXT:    xscvqpdp 2, [[REG4]]
+; CHECK-NEXT:    stxsd 2, 0(3)
 ; CHECK-NEXT:    blr
                                  fp128* nocapture %f2, i32 signext %i1, i8 zeroext %c1,
                                  <4 x i32>* nocapture %vec2) {