From 3dcb2a2d92cca5d5ebac11fafab923520341b265 Mon Sep 17 00:00:00 2001
From: Gautam Chakrabarti <gchakrabarti@nvidia.com>
Date: Tue, 28 Jan 2014 18:35:29 +0000
Subject: [PATCH] [NVPTX] Fix emitting aggregate parameters

The code was missing the case for aggregate parameters and
hence was emitting them as .b0 type. Also fixed a couple
of comments.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@200325 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/NVPTX/NVPTXAsmPrinter.cpp |  6 +++---
 test/CodeGen/NVPTX/aggr-param.ll     | 20 ++++++++++++++++++++
 2 files changed, 23 insertions(+), 3 deletions(-)
 create mode 100644 test/CodeGen/NVPTX/aggr-param.ll

diff --git a/lib/Target/NVPTX/NVPTXAsmPrinter.cpp b/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
index 2c8e7e70f68..5fad27e47a9 100644
--- a/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
+++ b/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
@@ -1522,8 +1522,8 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
     }
 
     if (PAL.hasAttribute(paramIndex + 1, Attribute::ByVal) == false) {
-      if (Ty->isVectorTy()) {
-        // Just print .param .b8 .align <a> .param[size];
+      if (Ty->isAggregateType() || Ty->isVectorTy()) {
+        // Just print .param .align <a> .b8 .param[size];
         // <a> = PAL.getparamalignment
         // size = typeallocsize of element type
         unsigned align = PAL.getParamAlignment(paramIndex + 1);
@@ -1603,7 +1603,7 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
     Type *ETy = PTy->getElementType();
 
     if (isABI || isKernelFunc) {
-      // Just print .param .b8 .align <a> .param[size];
+      // Just print .param .align <a> .b8 .param[size];
       // <a> = PAL.getparamalignment
       // size = typeallocsize of element type
       unsigned align = PAL.getParamAlignment(paramIndex + 1);
diff --git a/test/CodeGen/NVPTX/aggr-param.ll b/test/CodeGen/NVPTX/aggr-param.ll
new file mode 100644
index 00000000000..21deb7ebce8
--- /dev/null
+++ b/test/CodeGen/NVPTX/aggr-param.ll
@@ -0,0 +1,20 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
+
+; Make sure aggregate param types get emitted properly.
+
+%struct.float4 = type { float, float, float, float }
+
+; CHECK: .visible .func bar
+; CHECK:   .param .align 4 .b8 bar_param_0[16]
+define void @bar(%struct.float4 %f) {
+entry:
+  ret void
+}
+
+; CHECK: .visible .func foo
+; CHECK:   .param .align 4 .b8 foo_param_0[20]
+define void @foo([5 x i32] %f) {
+entry:
+  ret void
+}
+