From: Justin Holewinski <jholewinski@nvidia.com>
Date: Mon, 20 May 2013 12:13:28 +0000 (+0000)
Subject: [NVPTX] Fix i1 kernel parameters and global variables.  ABI rules say we need to... 
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=55fdf5362954a58b7e1bda34ff5c7e1246bfb991;p=oota-llvm.git

[NVPTX] Fix i1 kernel parameters and global variables.  ABI rules say we need to use .u8 for i1 parameters for kernels.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@182253 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/NVPTX/NVPTXAsmPrinter.cpp b/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
index ce5d78afa33..e2832b02b53 100644
--- a/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
+++ b/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
@@ -1199,7 +1199,11 @@ void NVPTXAsmPrinter::printModuleLevelGV(GlobalVariable *GVar, raw_ostream &O,
 
   if (ETy->isPrimitiveType() || ETy->isIntegerTy() || isa<PointerType>(ETy)) {
     O << " .";
-    O << getPTXFundamentalTypeStr(ETy, false);
+    // Special case: ABI requires that we use .u8 for predicates
+    if (ETy->isIntegerTy(1))
+      O << "u8";
+    else
+      O << getPTXFundamentalTypeStr(ETy, false);
     O << " ";
     O << *Mang->getSymbol(GVar);
 
@@ -1564,7 +1568,13 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
         }
 
         // non-pointer scalar to kernel func
-        O << "\t.param ." << getPTXFundamentalTypeStr(Ty) << " ";
+        O << "\t.param .";
+        // Special case: predicate operands become .u8 types
+        if (Ty->isIntegerTy(1))
+          O << "u8";
+        else
+          O << getPTXFundamentalTypeStr(Ty);
+        O << " ";
         printParamName(I, paramIndex, O);
         continue;
       }
diff --git a/test/CodeGen/NVPTX/i1-global.ll b/test/CodeGen/NVPTX/i1-global.ll
new file mode 100644
index 00000000000..0595325977e
--- /dev/null
+++ b/test/CodeGen/NVPTX/i1-global.ll
@@ -0,0 +1,19 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 -drvcuda | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64"
+
+
+; CHECK: .visible .global .align 1 .u8 mypred
+@mypred = addrspace(1) global i1 true, align 1
+
+
+define void @foo(i1 %p, i32* %out) {
+  %ld = load i1 addrspace(1)* @mypred
+  %val = zext i1 %ld to i32
+  store i32 %val, i32* %out
+  ret void
+}
+
+
+!nvvm.annotations = !{!0}
+!0 = metadata !{void (i1, i32*)* @foo, metadata !"kernel", i32 1}
diff --git a/test/CodeGen/NVPTX/i1-param.ll b/test/CodeGen/NVPTX/i1-param.ll
new file mode 100644
index 00000000000..fabd61a25d2
--- /dev/null
+++ b/test/CodeGen/NVPTX/i1-param.ll
@@ -0,0 +1,18 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 -drvcuda | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64"
+
+; Make sure predicate (i1) operands to kernels get expanded out to .u8
+
+; CHECK: .entry foo
+; CHECK:   .param .u8 foo_param_0
+; CHECK:   .param .u32 foo_param_1
+define void @foo(i1 %p, i32* %out) {
+  %val = zext i1 %p to i32
+  store i32 %val, i32* %out
+  ret void
+}
+
+
+!nvvm.annotations = !{!0}
+!0 = metadata !{void (i1, i32*)* @foo, metadata !"kernel", i32 1}