[llvm] r258536 - [NVPTX] expand mul_lohi to mul_lo and mul_hi

[Jingyue Wu via llvm-commits]

Author: jingyue
Date: Fri Jan 22 13:47:26 2016
New Revision: 258536

URL: http://llvm.org/viewvc/llvm-project?rev=258536&view=rev
Log:
[NVPTX] expand mul_lohi to mul_lo and mul_hi

Summary: Fixes PR26186.

Reviewers: grosser, jholewinski

Subscribers: jholewinski, llvm-commits

Differential Revision: http://reviews.llvm.org/D16479

Modified:
    llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.cpp
    llvm/trunk/test/CodeGen/NVPTX/arithmetic-int.ll

Modified: llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.cpp?rev=258536&r1=258535&r2=258536&view=diff
==============================================================================
--- llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.cpp Fri Jan 22 13:47:26 2016
@@ -273,6 +273,10 @@ NVPTXTargetLowering::NVPTXTargetLowering
   // PTX does not directly support SELP of i1, so promote to i32 first
   setOperationAction(ISD::SELECT, MVT::i1, Custom);
 
+  // PTX cannot multiply two i64s in a single instruction.
+  setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
+  setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
+
   // We have some custom DAG combine patterns for these nodes
   setTargetDAGCombine(ISD::ADD);
   setTargetDAGCombine(ISD::AND);

Modified: llvm/trunk/test/CodeGen/NVPTX/arithmetic-int.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/NVPTX/arithmetic-int.ll?rev=258536&r1=258535&r2=258536&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/NVPTX/arithmetic-int.ll (original)
+++ llvm/trunk/test/CodeGen/NVPTX/arithmetic-int.ll Fri Jan 22 13:47:26 2016
@@ -29,6 +29,30 @@ define i64 @mul_i64(i64 %a, i64 %b) {
   ret i64 %ret
 }
 
+define i64 @umul_lohi_i64(i64 %a) {
+; CHECK-LABEL: umul_lohi_i64(
+entry:
+  %0 = zext i64 %a to i128
+  %1 = mul i128 %0, 288
+; CHECK: mul.lo.{{u|s}}64
+; CHECK: mul.hi.{{u|s}}64
+  %2 = lshr i128 %1, 1
+  %3 = trunc i128 %2 to i64
+  ret i64 %3
+}
+
+define i64 @smul_lohi_i64(i64 %a) {
+; CHECK-LABEL: smul_lohi_i64(
+entry:
+  %0 = sext i64 %a to i128
+  %1 = mul i128 %0, 288
+; CHECK: mul.lo.{{u|s}}64
+; CHECK: mul.hi.{{u|s}}64
+  %2 = ashr i128 %1, 1
+  %3 = trunc i128 %2 to i64
+  ret i64 %3
+}
+
 define i64 @sdiv_i64(i64 %a, i64 %b) {
 ; CHECK: div.s64 %rd{{[0-9]+}}, %rd{{[0-9]+}}, %rd{{[0-9]+}}
 ; CHECK: ret