[llvm] fdda800 - AMDGPU: Fix opaque pointer and other bugs in printf of constant strings
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 10 10:39:49 PST 2023
Author: Matt Arsenault
Date: 2023-01-10T13:39:44-05:00
New Revision: fdda800ba35e435db45475c404b995de2d5ecbb8
URL: https://github.com/llvm/llvm-project/commit/fdda800ba35e435db45475c404b995de2d5ecbb8
DIFF: https://github.com/llvm/llvm-project/commit/fdda800ba35e435db45475c404b995de2d5ecbb8.diff
LOG: AMDGPU: Fix opaque pointer and other bugs in printf of constant strings
Strip pointer casts to get to the global. Fixes not respecting indexed
constant strings. Tolerate non-null terminated and empty strings.
Added:
Modified:
llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp
llvm/test/CodeGen/AMDGPU/opencl-printf.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp b/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp
index 5fd48307f490..d5c76967a68f 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp
@@ -22,6 +22,7 @@
#include "llvm/ADT/Triple.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/DiagnosticInfo.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/IRBuilder.h"
@@ -136,6 +137,17 @@ static bool shouldPrintAsStr(char Specifier, Type *OpType) {
return Specifier == 's' && isa<PointerType>(OpType);
}
+constexpr StringLiteral NonLiteralStr("???");
+static_assert(NonLiteralStr.size() == 3);
+
+static StringRef getAsConstantStr(Value *V) {
+ StringRef S;
+ if (!getConstantStringInfo(V, S))
+ S = NonLiteralStr;
+
+ return S;
+}
+
static void diagnoseInvalidFormatString(const CallBase *CI) {
DiagnosticInfoUnsupported UnsupportedFormatStr(
*CI->getParent()->getParent(),
@@ -150,8 +162,6 @@ bool AMDGPUPrintfRuntimeBindingImpl::lowerPrintfForGpu(Module &M) {
IRBuilder<> Builder(Ctx);
Type *I32Ty = Type::getInt32Ty(Ctx);
unsigned UniqID = 0;
- constexpr StringLiteral NonLiteralStr("???");
- static_assert(NonLiteralStr.size() == 3);
for (auto *CI : Printfs) {
unsigned NumOps = CI->arg_size();
@@ -246,31 +256,9 @@ bool AMDGPUPrintfRuntimeBindingImpl::lowerPrintfForGpu(Module &M) {
ArgSize = 4;
}
}
- if (shouldPrintAsStr(OpConvSpecifiers[ArgCount - 1], ArgType)) {
- ArgSize = NonLiteralStr.size() + 1;
- if (auto *ConstExpr = dyn_cast<ConstantExpr>(Arg)) {
- auto *GV = dyn_cast<GlobalVariable>(ConstExpr->getOperand(0));
- if (GV && GV->hasInitializer()) {
- Constant *Init = GV->getInitializer();
- bool IsZeroValue = Init->isZeroValue();
- auto *CA = dyn_cast<ConstantDataArray>(Init);
- if (IsZeroValue || (CA && CA->isString())) {
- size_t SizeStr =
- IsZeroValue ? 1 : (strlen(CA->getAsCString().data()) + 1);
- size_t Rem = SizeStr % DWORD_ALIGN;
- size_t NSizeStr = 0;
- LLVM_DEBUG(dbgs() << "Printf string original size = " << SizeStr
- << '\n');
- if (Rem) {
- NSizeStr = SizeStr + (DWORD_ALIGN - Rem);
- } else {
- NSizeStr = SizeStr;
- }
- ArgSize = NSizeStr;
- }
- }
- }
- }
+ if (shouldPrintAsStr(OpConvSpecifiers[ArgCount - 1], ArgType))
+ ArgSize = alignTo(getAsConstantStr(Arg).size() + 1, 4);
+
LLVM_DEBUG(dbgs() << "Printf ArgSize (in buffer) = " << ArgSize
<< " for type: " << *ArgType << '\n');
Sizes << ArgSize << ':';
@@ -412,21 +400,9 @@ bool AMDGPUPrintfRuntimeBindingImpl::lowerPrintfForGpu(Module &M) {
}
Arg = new BitCastInst(Arg, IType, "PrintArgFP", Brnch);
WhatToStore.push_back(Arg);
- } else if (ArgType->getTypeID() == Type::PointerTyID) {
+ } else if (isa<PointerType>(ArgType)) {
if (shouldPrintAsStr(OpConvSpecifiers[ArgCount - 1], ArgType)) {
- StringRef S = NonLiteralStr;
- if (auto *ConstExpr = dyn_cast<ConstantExpr>(Arg)) {
- auto *GV = dyn_cast<GlobalVariable>(ConstExpr->getOperand(0));
- if (GV && GV->hasInitializer()) {
- Constant *Init = GV->getInitializer();
- bool IsZeroValue = Init->isZeroValue();
- auto *CA = dyn_cast<ConstantDataArray>(Init);
- if (IsZeroValue || (CA && CA->isString())) {
- S = IsZeroValue ? "" : CA->getAsCString();
- }
- }
- }
-
+ StringRef S = getAsConstantStr(Arg);
if (!S.empty()) {
const uint64_t ReadSize = 4;
@@ -459,7 +435,7 @@ bool AMDGPUPrintfRuntimeBindingImpl::lowerPrintfForGpu(Module &M) {
} else {
WhatToStore.push_back(Arg);
}
- } else {
+ } else {
WhatToStore.push_back(Arg);
}
for (unsigned I = 0, E = WhatToStore.size(); I != E; ++I) {
diff --git a/llvm/test/CodeGen/AMDGPU/opencl-printf.ll b/llvm/test/CodeGen/AMDGPU/opencl-printf.ll
index e2f673af3ba5..a7e5289e527c 100644
--- a/llvm/test/CodeGen/AMDGPU/opencl-printf.ll
+++ b/llvm/test/CodeGen/AMDGPU/opencl-printf.ll
@@ -873,7 +873,7 @@ define amdgpu_kernel void @test_print_string_literal_size0(i32 %n) {
; GCN-NEXT: store i32 17, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -905,7 +905,7 @@ define amdgpu_kernel void @test_print_string_literal_1ai8.zero(i32 %n) {
; GCN-NEXT: store i32 18, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -937,7 +937,7 @@ define amdgpu_kernel void @test_print_string_literal_1ai8.undef(i32 %n) {
; GCN-NEXT: store i32 19, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -969,7 +969,7 @@ define amdgpu_kernel void @test_print_string_literal_i8.zero(i32 %n) {
; GCN-NEXT: store i32 20, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1001,7 +1001,7 @@ define amdgpu_kernel void @test_print_string_literal_size2(i32 %n) {
; GCN-NEXT: store i32 21, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 27, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1033,7 +1033,7 @@ define amdgpu_kernel void @test_print_string_literal_size3(i32 %n) {
; GCN-NEXT: store i32 22, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 7195, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1065,7 +1065,7 @@ define amdgpu_kernel void @test_print_string_literal_size3_zero(i32 %n) {
; GCN-NEXT: store i32 23, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1097,7 +1097,7 @@ define amdgpu_kernel void @test_print_string_literal_size4(i32 %n) {
; GCN-NEXT: store i32 24, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 6513249, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1118,7 +1118,7 @@ define amdgpu_kernel void @test_print_string_literal_size4_nonull_term(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size4_nonull_term(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 16)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1129,7 +1129,7 @@ define amdgpu_kernel void @test_print_string_literal_size4_nonull_term(i32 %n) {
; GCN-NEXT: store i32 25, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1150,7 +1150,7 @@ define amdgpu_kernel void @test_print_string_literal_size5(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size5(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 16)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1161,7 +1161,7 @@ define amdgpu_kernel void @test_print_string_literal_size5(i32 %n) {
; GCN-NEXT: store i32 26, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1182,7 +1182,7 @@ define amdgpu_kernel void @test_print_string_literal_size6(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size6(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 16)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1193,10 +1193,13 @@ define amdgpu_kernel void @test_print_string_literal_size6(i32 %n) {
; GCN-NEXT: store i32 27, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 101, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1214,7 +1217,7 @@ define amdgpu_kernel void @test_print_string_literal_size7(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size7(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 16)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1225,10 +1228,13 @@ define amdgpu_kernel void @test_print_string_literal_size7(i32 %n) {
; GCN-NEXT: store i32 28, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 101, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1246,7 +1252,7 @@ define amdgpu_kernel void @test_print_string_literal_size8(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size8(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 16)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1257,10 +1263,13 @@ define amdgpu_kernel void @test_print_string_literal_size8(i32 %n) {
; GCN-NEXT: store i32 29, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 7565921, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1278,7 +1287,7 @@ define amdgpu_kernel void @test_print_string_literal_size9(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size9(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 20)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1289,10 +1298,13 @@ define amdgpu_kernel void @test_print_string_literal_size9(i32 %n) {
; GCN-NEXT: store i32 30, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1684104548, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 1717921122, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1310,7 +1322,7 @@ define amdgpu_kernel void @test_print_string_literal_size16(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size16(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 24)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1321,10 +1333,19 @@ define amdgpu_kernel void @test_print_string_literal_size16(i32 %n) {
; GCN-NEXT: store i32 31, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 909194801, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 909194801, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
+; GCN-NEXT: store i32 909194801, ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST5:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR4]] to ptr addrspace(1)
+; GCN-NEXT: store i32 3225137, ptr addrspace(1) [[PRINTBUFFPTRCAST5]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR6:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST7:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR6]] to ptr addrspace(1)
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST7]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1342,7 +1363,7 @@ define amdgpu_kernel void @test_print_string_literal_size17(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size17(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 28)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1353,10 +1374,19 @@ define amdgpu_kernel void @test_print_string_literal_size17(i32 %n) {
; GCN-NEXT: store i32 32, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 925972273, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 925972273, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
+; GCN-NEXT: store i32 925972273, ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST5:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR4]] to ptr addrspace(1)
+; GCN-NEXT: store i32 925972273, ptr addrspace(1) [[PRINTBUFFPTRCAST5]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR6:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST7:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR6]] to ptr addrspace(1)
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST7]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1374,7 +1404,7 @@ define amdgpu_kernel void @test_print_string_literal_size20(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size20(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 28)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1385,10 +1415,22 @@ define amdgpu_kernel void @test_print_string_literal_size20(i32 %n) {
; GCN-NEXT: store i32 33, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 1751606885, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
+; GCN-NEXT: store i32 1818978921, ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST5:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR4]] to ptr addrspace(1)
+; GCN-NEXT: store i32 1886350957, ptr addrspace(1) [[PRINTBUFFPTRCAST5]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR6:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST7:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR6]] to ptr addrspace(1)
+; GCN-NEXT: store i32 7565937, ptr addrspace(1) [[PRINTBUFFPTRCAST7]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR8:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR6]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST9:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR8]] to ptr addrspace(1)
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST9]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1406,7 +1448,7 @@ define amdgpu_kernel void @test_print_string_literal_size32(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_size32(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 40)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1417,10 +1459,31 @@ define amdgpu_kernel void @test_print_string_literal_size32(i32 %n) {
; GCN-NEXT: store i32 34, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
+; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST5:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR4]] to ptr addrspace(1)
+; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST5]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR6:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST7:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR6]] to ptr addrspace(1)
+; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST7]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR8:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR6]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST9:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR8]] to ptr addrspace(1)
+; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST9]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR10:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR8]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST11:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR10]] to ptr addrspace(1)
+; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST11]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR12:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR10]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST13:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR12]] to ptr addrspace(1)
+; GCN-NEXT: store i32 7565921, ptr addrspace(1) [[PRINTBUFFPTRCAST13]], align 4
+; GCN-NEXT: [[PRINTBUFFNEXTPTR14:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR12]], i32 4
+; GCN-NEXT: [[PRINTBUFFPTRCAST15:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR14]] to ptr addrspace(1)
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST15]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1502,7 +1565,7 @@ define amdgpu_kernel void @test_print_string_literal_v4i8(i32 %n) {
;
; GCN-LABEL: @test_print_string_literal_v4i8(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 16)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1513,7 +1576,7 @@ define amdgpu_kernel void @test_print_string_literal_v4i8(i32 %n) {
; GCN-NEXT: store i32 37, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 67305985, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1545,7 +1608,7 @@ define amdgpu_kernel void @test_print_string_literal_v4i32(i32 %n) {
; GCN-NEXT: store i32 38, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1577,7 +1640,7 @@ define amdgpu_kernel void @test_print_string_literal_struct(i32 %n) {
; GCN-NEXT: store i32 39, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 66, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1737,7 +1800,7 @@ define amdgpu_kernel void @test_print_string_float_neg0(i32 %n) {
; GCN-NEXT: store i32 44, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1769,7 +1832,7 @@ define amdgpu_kernel void @test_print_string_float_0(i32 %n) {
; GCN-NEXT: store i32 45, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1801,7 +1864,7 @@ define amdgpu_kernel void @test_print_string_ptr_null(i32 %n) {
; GCN-NEXT: store i32 46, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1833,7 +1896,7 @@ define amdgpu_kernel void @test_print_string_ptr_undef(i32 %n) {
; GCN-NEXT: store i32 47, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
@@ -1854,7 +1917,7 @@ define amdgpu_kernel void @test_print_string_indexed(i32 %n) {
;
; GCN-LABEL: @test_print_string_indexed(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 40)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 28)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1865,31 +1928,19 @@ define amdgpu_kernel void @test_print_string_indexed(i32 %n) {
; GCN-NEXT: store i32 48, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 1936875892, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
+; GCN-NEXT: store i32 1936875892, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
+; GCN-NEXT: store i32 1936875892, ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST5:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR4]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST5]], align 4
+; GCN-NEXT: store i32 1936875892, ptr addrspace(1) [[PRINTBUFFPTRCAST5]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR6:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST7:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR6]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST7]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR8:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR6]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST9:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR8]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST9]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR10:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR8]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST11:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR10]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST11]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR12:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR10]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST13:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR12]] to ptr addrspace(1)
-; GCN-NEXT: store i32 7565921, ptr addrspace(1) [[PRINTBUFFPTRCAST13]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR14:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR12]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST15:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR14]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST15]], align 4
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST7]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
@@ -1907,7 +1958,7 @@ define amdgpu_kernel void @test_print_string_indexed_oob(i32 %n) {
;
; GCN-LABEL: @test_print_string_indexed_oob(
; GCN-NEXT: entry:
-; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 40)
+; GCN-NEXT: [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 12)
; GCN-NEXT: br label [[ENTRY_SPLIT:%.*]]
; GCN: entry.split:
; GCN-NEXT: [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
@@ -1918,31 +1969,10 @@ define amdgpu_kernel void @test_print_string_indexed_oob(i32 %n) {
; GCN-NEXT: store i32 49, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
; GCN-NEXT: [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFGEP]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT: store i32 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
; GCN-NEXT: [[PRINTBUFFNEXTPTR:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFGEP]], i32 4
; GCN-NEXT: [[PRINTBUFFPTRCAST1:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST3:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR2]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST3]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST5:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR4]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST5]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR6:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST7:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR6]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST7]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR8:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR6]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST9:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR8]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST9]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR10:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR8]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST11:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR10]] to ptr addrspace(1)
-; GCN-NEXT: store i32 1953722977, ptr addrspace(1) [[PRINTBUFFPTRCAST11]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR12:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR10]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST13:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR12]] to ptr addrspace(1)
-; GCN-NEXT: store i32 7565921, ptr addrspace(1) [[PRINTBUFFPTRCAST13]], align 4
-; GCN-NEXT: [[PRINTBUFFNEXTPTR14:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR12]], i32 4
-; GCN-NEXT: [[PRINTBUFFPTRCAST15:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFNEXTPTR14]] to ptr addrspace(1)
-; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST15]], align 4
+; GCN-NEXT: store i32 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST1]], align 4
; GCN-NEXT: br label [[TMP2]]
; GCN: 2:
; GCN-NEXT: ret void
More information about the llvm-commits
mailing list