[llvm] r213276 - [NVPTX] Add missing .v4 qualifier on vector store instruction

Justin Holewinski jholewinski at nvidia.com
Thu Jul 17 09:58:56 PDT 2014


Author: jholewinski
Date: Thu Jul 17 11:58:56 2014
New Revision: 213276

URL: http://llvm.org/viewvc/llvm-project?rev=213276&view=rev
Log:
[NVPTX] Add missing .v4 qualifier on vector store instruction

Added:
    llvm/trunk/test/CodeGen/NVPTX/vector-call.ll
Modified:
    llvm/trunk/lib/Target/NVPTX/NVPTXInstrInfo.td

Modified: llvm/trunk/lib/Target/NVPTX/NVPTXInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/NVPTX/NVPTXInstrInfo.td?rev=213276&r1=213275&r2=213276&view=diff
==============================================================================
--- llvm/trunk/lib/Target/NVPTX/NVPTXInstrInfo.td (original)
+++ llvm/trunk/lib/Target/NVPTX/NVPTXInstrInfo.td Thu Jul 17 11:58:56 2014
@@ -1917,7 +1917,7 @@ def StoreParamV2I8   : StoreParamV2Inst<
 def StoreParamV4I32    : NVPTXInst<(outs), (ins Int32Regs:$val, Int32Regs:$val2,
                                                Int32Regs:$val3, Int32Regs:$val4,
                                                 i32imm:$a, i32imm:$b),
-                   "st.param.b32\t[param$a+$b], {{$val, $val2, $val3, $val4}};",
+                "st.param.v4.b32\t[param$a+$b], {{$val, $val2, $val3, $val4}};",
                          []>;
 
 def StoreParamV4I16    : NVPTXInst<(outs), (ins Int16Regs:$val, Int16Regs:$val2,

Added: llvm/trunk/test/CodeGen/NVPTX/vector-call.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/NVPTX/vector-call.ll?rev=213276&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/NVPTX/vector-call.ll (added)
+++ llvm/trunk/test/CodeGen/NVPTX/vector-call.ll Thu Jul 17 11:58:56 2014
@@ -0,0 +1,12 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
+
+target triple = "nvptx-unknown-cuda"
+
+declare void @bar(<4 x i32>)
+
+; CHECK-LABEL @foo
+define void @foo(<4 x i32> %a) {
+; CHECK: st.param.v4.b32
+  tail call void @bar(<4 x i32> %a)
+  ret void
+}





More information about the llvm-commits mailing list