[llvm] r330278 - [Power9]Legalize and emit code for converting Unsigned HWord/Char to Quad-Precision
Lei Huang via llvm-commits
llvm-commits at lists.llvm.org
Wed Apr 18 10:41:46 PDT 2018
Author: lei
Date: Wed Apr 18 10:41:46 2018
New Revision: 330278
URL: http://llvm.org/viewvc/llvm-project?rev=330278&view=rev
Log:
[Power9]Legalize and emit code for converting Unsigned HWord/Char to Quad-Precision
Legalize and emit code for converting unsigned HWord/Char to QP:
xscvsdqp
xscvudqp
Only covering patterns for unsigned forms cause we don't have part-word
sign-extending integer loads into VSX registers.
Differential Revision: https://reviews.llvm.org/D45494
Modified:
llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td
llvm/trunk/test/CodeGen/PowerPC/f128-conv.ll
Modified: llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td?rev=330278&r1=330277&r2=330278&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td (original)
+++ llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td Wed Apr 18 10:41:46 2018
@@ -3142,6 +3142,14 @@ let AddedComplexity = 400, Predicates =
def : Pat<(f128 (uint_to_fp (i64 (load ixaddr:$src)))),
(f128 (XSCVUDQP (LXSD ixaddr:$src)))>;
+ // Convert Unsigned HWord in memory -> QP
+ def : Pat<(f128 (uint_to_fp ScalarLoads.ZELi16)),
+ (f128 (XSCVUDQP (LXSIHZX xaddr:$src)))>;
+
+ // Convert Unsigned Byte in memory -> QP
+ def : Pat<(f128 (uint_to_fp ScalarLoads.ZELi8)),
+ (f128 (XSCVUDQP (LXSIBZX xoaddr:$src)))>;
+
} // end HasP9Vector, AddedComplexity
let Predicates = [HasP9Vector] in {
Modified: llvm/trunk/test/CodeGen/PowerPC/f128-conv.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/f128-conv.ll?rev=330278&r1=330277&r2=330278&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/f128-conv.ll (original)
+++ llvm/trunk/test/CodeGen/PowerPC/f128-conv.ll Wed Apr 18 10:41:46 2018
@@ -6,6 +6,8 @@
@umem = global [5 x i64] [i64 560, i64 100, i64 34, i64 2, i64 5], align 8
@swMem = global [5 x i32] [i32 5, i32 2, i32 3, i32 4, i32 0], align 4
@uwMem = global [5 x i32] [i32 5, i32 2, i32 3, i32 4, i32 0], align 4
+ at uhwMem = local_unnamed_addr global [5 x i16] [i16 5, i16 2, i16 3, i16 4, i16 0], align 2
+ at ubMem = local_unnamed_addr global [5 x i8] c"\05\02\03\04\00", align 1
; Function Attrs: norecurse nounwind
define void @sdwConv2qp(fp128* nocapture %a, i64 %b) {
@@ -297,3 +299,142 @@ entry:
; CHECK-NEXT: stxv [[CONV]], 0(3)
; CHECK-NEXT: blr
}
+
+; Function Attrs: norecurse nounwind
+define void @uhwConv2qp(fp128* nocapture %a, i16 zeroext %b) {
+entry:
+ %conv = uitofp i16 %b to fp128
+ store fp128 %conv, fp128* %a, align 16
+ ret void
+
+
+; CHECK-LABEL: uhwConv2qp
+; CHECK: mtvsrwz [[REG:[0-9]+]], 4
+; CHECK-NEXT: xscvudqp [[CONV:[0-9]+]], [[REG]]
+; CHECK-NEXT: stxv [[CONV]], 0(3)
+; CHECK-NEXT: blr
+}
+
+; Function Attrs: norecurse nounwind
+define void @uhwConv2qp_02(fp128* nocapture %a, i16* nocapture readonly %b) {
+entry:
+ %0 = load i16, i16* %b, align 2
+ %conv = uitofp i16 %0 to fp128
+ store fp128 %conv, fp128* %a, align 16
+ ret void
+
+; CHECK-LABEL: uhwConv2qp_02
+; CHECK: lxsihzx [[REG:[0-9]+]], 0, 4
+; CHECK-NEXT: xscvudqp [[CONV:[0-9]+]], [[REG]]
+; CHECK-NEXT: stxv [[CONV]], 0(3)
+; CHECK-NEXT: blr
+}
+
+; Function Attrs: norecurse nounwind
+define void @uhwConv2qp_03(fp128* nocapture %a) {
+entry:
+ %0 = load i16, i16* getelementptr inbounds
+ ([5 x i16], [5 x i16]* @uhwMem, i64 0, i64 3), align 2
+ %conv = uitofp i16 %0 to fp128
+ store fp128 %conv, fp128* %a, align 16
+ ret void
+
+; CHECK-LABEL: uhwConv2qp_03
+; CHECK: addis [[REG0:[0-9]+]], 2, .LC4 at toc@ha
+; CHECK: ld [[REG0]], .LC4 at toc@l([[REG0]])
+; CHECK: addi [[REG0]], [[REG0]], 6
+; CHECK: lxsihzx [[REG:[0-9]+]], 0, [[REG0]]
+; CHECK-NEXT: xscvudqp [[CONV:[0-9]+]], [[REG]]
+; CHECK-NEXT: stxv [[CONV]], 0(3)
+; CHECK-NEXT: blr
+}
+
+; Function Attrs: norecurse nounwind
+define void @uhwConv2qp_04(fp128* nocapture %a, i16 zeroext %b,
+ i16* nocapture readonly %c) {
+entry:
+ %conv = zext i16 %b to i32
+ %0 = load i16, i16* %c, align 2
+ %conv1 = zext i16 %0 to i32
+ %add = add nuw nsw i32 %conv1, %conv
+ %conv2 = sitofp i32 %add to fp128
+ store fp128 %conv2, fp128* %a, align 16
+ ret void
+
+; CHECK-LABEL: uhwConv2qp_04
+; CHECK: lhz [[REG0:[0-9]+]], 0(5)
+; CHECK: add 4, [[REG0]], 4
+; CHECK: mtvsrwa [[REG:[0-9]+]], 4
+; CHECK-NEXT: xscvsdqp [[CONV:[0-9]+]], [[REG]]
+; CHECK-NEXT: stxv [[CONV]], 0(3)
+; CHECK-NEXT: blr
+}
+
+; Function Attrs: norecurse nounwind
+define void @ubConv2qp(fp128* nocapture %a, i8 zeroext %b) {
+entry:
+ %conv = uitofp i8 %b to fp128
+ store fp128 %conv, fp128* %a, align 16
+ ret void
+
+; CHECK-LABEL: ubConv2qp
+; CHECK: mtvsrwz [[REG:[0-9]+]], 4
+; CHECK-NEXT: xscvudqp [[CONV:[0-9]+]], [[REG]]
+; CHECK-NEXT: stxv [[CONV]], 0(3)
+; CHECK-NEXT: blr
+}
+
+; Function Attrs: norecurse nounwind
+define void @ubConv2qp_02(fp128* nocapture %a, i8* nocapture readonly %b) {
+entry:
+ %0 = load i8, i8* %b, align 1
+ %conv = uitofp i8 %0 to fp128
+ store fp128 %conv, fp128* %a, align 16
+ ret void
+
+; CHECK-LABEL: ubConv2qp_02
+; CHECK: lxsibzx [[REG:[0-9]+]], 0, 4
+; CHECK-NEXT: xscvudqp [[CONV:[0-9]+]], [[REG]]
+; CHECK-NEXT: stxv [[CONV]], 0(3)
+; CHECK-NEXT: blr
+}
+
+; Function Attrs: norecurse nounwind
+define void @ubConv2qp_03(fp128* nocapture %a) {
+entry:
+ %0 = load i8, i8* getelementptr inbounds
+ ([5 x i8], [5 x i8]* @ubMem, i64 0, i64 2), align 1
+ %conv = uitofp i8 %0 to fp128
+ store fp128 %conv, fp128* %a, align 16
+ ret void
+
+; CHECK-LABEL: ubConv2qp_03
+; CHECK: addis [[REG0:[0-9]+]], 2, .LC5 at toc@ha
+; CHECK: ld [[REG0]], .LC5 at toc@l([[REG0]])
+; CHECK: addi [[REG0]], [[REG0]], 2
+; CHECK: lxsibzx [[REG:[0-9]+]], 0, [[REG0]]
+; CHECK-NEXT: xscvudqp [[CONV:[0-9]+]], [[REG]]
+; CHECK-NEXT: stxv [[CONV]], 0(3)
+; CHECK-NEXT: blr
+}
+
+; Function Attrs: norecurse nounwind
+define void @ubConv2qp_04(fp128* nocapture %a, i8 zeroext %b,
+ i8* nocapture readonly %c) {
+entry:
+ %conv = zext i8 %b to i32
+ %0 = load i8, i8* %c, align 1
+ %conv1 = zext i8 %0 to i32
+ %add = add nuw nsw i32 %conv1, %conv
+ %conv2 = sitofp i32 %add to fp128
+ store fp128 %conv2, fp128* %a, align 16
+ ret void
+
+; CHECK-LABEL: ubConv2qp_04
+; CHECK: lbz [[REG0:[0-9]+]], 0(5)
+; CHECK: add 4, [[REG0]], 4
+; CHECK: mtvsrwa [[REG:[0-9]+]], 4
+; CHECK-NEXT: xscvsdqp [[CONV:[0-9]+]], [[REG]]
+; CHECK-NEXT: stxv [[CONV]], 0(3)
+; CHECK-NEXT: blr
+}
More information about the llvm-commits
mailing list