From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Fri, 6 Jan 2023 19:55:47 +0000 (-0500)
Subject: AMDGPU: Use getConstantStringInfo for printf format strings
X-Git-Tag: upstream/17.0.6~20983
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=0d87732a1fccc111baf241a78c22634e7f3880e8;p=platform%2Fupstream%2Fllvm.git

AMDGPU: Use getConstantStringInfo for printf format strings

Tolerated printf format strings that are indexed globals and fixes
asserting on non-null terminated strings.
---

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp b/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp
index 50075ff..b243009 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPrintfRuntimeBinding.cpp
@@ -192,31 +192,17 @@ bool AMDGPUPrintfRuntimeBindingImpl::lowerPrintfForGpu(Module &M) {
         Op = Op_simplified;
     }
 
-    Value *Stripped = Op->stripPointerCasts();
-    if (isa<UndefValue>(Stripped) || isa<ConstantPointerNull>(Stripped))
-      continue;
-
-    GlobalVariable *GVar = dyn_cast<GlobalVariable>(Stripped);
-    if (!GVar || !GVar->hasDefinitiveInitializer() || !GVar->isConstant()) {
-      diagnoseInvalidFormatString(CI);
-      continue;
-    }
-
-    auto *Init = GVar->getInitializer();
-    if (isa<UndefValue>(Init) || isa<ConstantAggregateZero>(Init))
-      continue;
-
-    auto *CA = dyn_cast<ConstantDataArray>(Init);
-    if (!CA || !CA->isString()) {
-      diagnoseInvalidFormatString(CI);
+    StringRef FormatStr;
+    if (!getConstantStringInfo(Op, FormatStr)) {
+      Value *Stripped = Op->stripPointerCasts();
+      if (!isa<UndefValue>(Stripped) && !isa<ConstantPointerNull>(Stripped))
+        diagnoseInvalidFormatString(CI);
       continue;
     }
 
-    StringRef Str(CA->getAsCString());
-
     // We need this call to ascertain that we are printing a string or a
     // pointer. It takes out the specifiers and fills up the first arg.
-    getConversionSpecifiers(OpConvSpecifiers, Str, NumOps - 1);
+    getConversionSpecifiers(OpConvSpecifiers, FormatStr, NumOps - 1);
 
     // Add metadata for the string
     std::string AStreamHolder;
@@ -277,9 +263,9 @@ bool AMDGPUPrintfRuntimeBindingImpl::lowerPrintfForGpu(Module &M) {
       Sizes << ArgSize << ':';
       Sum += ArgSize;
     }
-    LLVM_DEBUG(dbgs() << "Printf format string in source = " << Str.str()
+    LLVM_DEBUG(dbgs() << "Printf format string in source = " << FormatStr
                       << '\n');
-    for (char C : Str) {
+    for (char C : FormatStr) {
       // Rest of the C escape sequences (e.g. \') are handled correctly
       // by the MDParser
       switch (C) {
diff --git a/llvm/test/CodeGen/AMDGPU/opencl-printf-unsupported.ll b/llvm/test/CodeGen/AMDGPU/opencl-printf-unsupported.ll
index 805bf0d..51c0336 100644
--- a/llvm/test/CodeGen/AMDGPU/opencl-printf-unsupported.ll
+++ b/llvm/test/CodeGen/AMDGPU/opencl-printf-unsupported.ll
@@ -59,24 +59,6 @@ entry:
   ret void
 }
 
-@struct = private unnamed_addr addrspace(4) constant { [6 x i8] } { [6 x i8] [i8 1, i8 2, i8 3, i8 4, i8 5, i8 6] }
-
-; CHECK: error: <unknown>:0:0: in function test_format_struct void (i32): printf format string must be a trivially resolved constant string global variable
-define amdgpu_kernel void @test_format_struct(i32 %n) {
-  %str = alloca [9 x i8], align 1, addrspace(5)
-  %call1 = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) @struct, ptr addrspace(5) %str, i32 %n)
-  ret void
-}
-
-@array.i16 = private unnamed_addr addrspace(4) constant [6 x i16] [i16 1, i16 2, i16 3, i16 4, i16 5, i16 6]
-
-; CHECK: error: <unknown>:0:0: in function test_format_array_i16 void (i32): printf format string must be a trivially resolved constant string global variable
-define amdgpu_kernel void @test_format_array_i16(i32 %n) {
-  %str = alloca [9 x i8], align 1, addrspace(5)
-  %call1 = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) @array.i16, ptr addrspace(5) %str, i32 %n)
-  ret void
-}
-
 declare i32 @printf(ptr addrspace(4), ...)
 
 !llvm.dbg.cu = !{!0}
diff --git a/llvm/test/CodeGen/AMDGPU/opencl-printf.ll b/llvm/test/CodeGen/AMDGPU/opencl-printf.ll
index 492cc75..f4631d1 100644
--- a/llvm/test/CodeGen/AMDGPU/opencl-printf.ll
+++ b/llvm/test/CodeGen/AMDGPU/opencl-printf.ll
@@ -4,6 +4,11 @@
 ; RUN: opt -mtriple=amdgcn--amdhsa -passes=amdgpu-printf-runtime-binding -mcpu=fiji -S < %s | FileCheck --check-prefix=GCN %s
 
 @.str = private unnamed_addr addrspace(4) constant [6 x i8] c"%s:%d\00", align 1
+
+@format.str.no.null.terminator = private unnamed_addr addrspace(4) constant [10 x i8] c"abcdefg %d", align 1
+@indexed.format.str = private unnamed_addr addrspace(4) constant [11 x i8] c"abcdefg %d\00", align 1
+
+
 @printed.str.size0 = private addrspace(4) constant [0 x i8] zeroinitializer, align 1
 @printed.str.1ai8.zero = private addrspace(4) constant [1 x i8] zeroinitializer, align 1
 @printed.str.1ai8.undef = private addrspace(4) constant [1 x i8] undef, align 1
@@ -598,6 +603,92 @@ entry:
   ret void
 }
 
+define amdgpu_kernel void @test_format_str_no_null_terminator(i32 %n) {
+; R600-LABEL: @test_format_str_no_null_terminator(
+; R600-NEXT:  entry:
+; R600-NEXT:    [[CALL1:%.*]] = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) @format.str.no.null.terminator, i32 [[N:%.*]])
+; R600-NEXT:    ret void
+;
+; GCN-LABEL: @test_format_str_no_null_terminator(
+; GCN-NEXT:  entry:
+; GCN-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 8)
+; GCN-NEXT:    br label [[ENTRY_SPLIT:%.*]]
+; GCN:       entry.split:
+; GCN-NEXT:    [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
+; GCN-NEXT:    br i1 [[TMP0]], label [[TMP1:%.*]], label [[TMP2:%.*]]
+; GCN:       1:
+; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
+; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
+; GCN-NEXT:    store i32 12, 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 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT:    br label [[TMP2]]
+; GCN:       2:
+; GCN-NEXT:    ret void
+;
+entry:
+  %call1 = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) @format.str.no.null.terminator, i32 %n)
+  ret void
+}
+
+
+define amdgpu_kernel void @test_indexed_format_str(i32 %n) {
+; R600-LABEL: @test_indexed_format_str(
+; R600-NEXT:  entry:
+; R600-NEXT:    [[CALL1:%.*]] = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) getelementptr inbounds ([11 x i8], ptr addrspace(4) @indexed.format.str, i64 0, i32 7), i32 [[N:%.*]])
+; R600-NEXT:    ret void
+;
+; GCN-LABEL: @test_indexed_format_str(
+; GCN-NEXT:  entry:
+; GCN-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 8)
+; GCN-NEXT:    br label [[ENTRY_SPLIT:%.*]]
+; GCN:       entry.split:
+; GCN-NEXT:    [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
+; GCN-NEXT:    br i1 [[TMP0]], label [[TMP1:%.*]], label [[TMP2:%.*]]
+; GCN:       1:
+; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
+; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
+; GCN-NEXT:    store i32 13, 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 [[N:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
+; GCN-NEXT:    br label [[TMP2]]
+; GCN:       2:
+; GCN-NEXT:    ret void
+;
+entry:
+  %call1 = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) getelementptr ([11 x i8], ptr addrspace(4) @indexed.format.str, i64 0, i32 7), i32 %n)
+  ret void
+}
+
+define amdgpu_kernel void @test_indexed_format_str_oob(i32 %n) {
+; R600-LABEL: @test_indexed_format_str_oob(
+; R600-NEXT:  entry:
+; R600-NEXT:    [[CALL1:%.*]] = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) getelementptr inbounds ([11 x i8], ptr addrspace(4) @indexed.format.str, i64 1, i64 0), i32 [[N:%.*]])
+; R600-NEXT:    ret void
+;
+; GCN-LABEL: @test_indexed_format_str_oob(
+; GCN-NEXT:  entry:
+; GCN-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 4)
+; GCN-NEXT:    br label [[ENTRY_SPLIT:%.*]]
+; GCN:       entry.split:
+; GCN-NEXT:    [[TMP0:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
+; GCN-NEXT:    br i1 [[TMP0]], label [[TMP1:%.*]], label [[TMP2:%.*]]
+; GCN:       1:
+; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
+; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
+; GCN-NEXT:    store i32 14, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
+; GCN-NEXT:    br label [[TMP2]]
+; GCN:       2:
+; GCN-NEXT:    ret void
+;
+entry:
+  %call1 = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) getelementptr ([11 x i8], ptr addrspace(4) @indexed.format.str, i64 0, i64 11), i32 %n)
+  ret void
+}
+
 define amdgpu_kernel void @string_pointee_type(i32 %n) {
 ; R600-LABEL: @string_pointee_type(
 ; R600-NEXT:    [[STR:%.*]] = alloca [9 x i8], align 1, addrspace(5)
@@ -614,7 +705,7 @@ define amdgpu_kernel void @string_pointee_type(i32 %n) {
 ; GCN:       2:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 12, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 15, 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
@@ -644,7 +735,7 @@ define amdgpu_kernel void @string_address_space4(i32 %n, ptr addrspace(4) %str)
 ; GCN:       2:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 13, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 16, 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
@@ -673,7 +764,7 @@ define amdgpu_kernel void @string_address_space1(i32 %n, ptr addrspace(1) %str)
 ; GCN:       2:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 14, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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
@@ -702,7 +793,7 @@ define amdgpu_kernel void @string_format_passed_i32(i32 %n, i32 %str) {
 ; GCN:       2:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 15, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 [[STR:%.*]], ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -740,7 +831,7 @@ define amdgpu_kernel void @test_kernel_addrspacecasted_format_str(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 16, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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
@@ -813,6 +904,18 @@ define amdgpu_kernel void @undef_initializer_gv(i32 %n) {
 ;
 ; GCN-LABEL: @undef_initializer_gv(
 ; GCN-NEXT:    [[STR:%.*]] = alloca [9 x i8], align 1, addrspace(5)
+; GCN-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 4)
+; GCN-NEXT:    br label [[DOTSPLIT:%.*]]
+; GCN:       .split:
+; GCN-NEXT:    [[TMP1:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
+; GCN-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]]
+; GCN:       2:
+; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
+; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
+; 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:    br label [[TMP3]]
+; GCN:       3:
 ; GCN-NEXT:    ret void
 ;
   %str = alloca [9 x i8], align 1, addrspace(5)
@@ -830,6 +933,18 @@ define amdgpu_kernel void @poison_initializer_gv(i32 %n) {
 ;
 ; GCN-LABEL: @poison_initializer_gv(
 ; GCN-NEXT:    [[STR:%.*]] = alloca [9 x i8], align 1, addrspace(5)
+; GCN-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 4)
+; GCN-NEXT:    br label [[DOTSPLIT:%.*]]
+; GCN:       .split:
+; GCN-NEXT:    [[TMP1:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
+; GCN-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]]
+; GCN:       2:
+; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
+; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
+; 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:    br label [[TMP3]]
+; GCN:       3:
 ; GCN-NEXT:    ret void
 ;
   %str = alloca [9 x i8], align 1, addrspace(5)
@@ -847,6 +962,18 @@ define amdgpu_kernel void @zero_initializer_gv(i32 %n) {
 ;
 ; GCN-LABEL: @zero_initializer_gv(
 ; GCN-NEXT:    [[STR:%.*]] = alloca [9 x i8], align 1, addrspace(5)
+; GCN-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 4)
+; GCN-NEXT:    br label [[DOTSPLIT:%.*]]
+; GCN:       .split:
+; GCN-NEXT:    [[TMP1:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
+; GCN-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]]
+; GCN:       2:
+; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
+; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
+; 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:    br label [[TMP3]]
+; GCN:       3:
 ; GCN-NEXT:    ret void
 ;
   %str = alloca [9 x i8], align 1, addrspace(5)
@@ -870,7 +997,7 @@ define amdgpu_kernel void @test_print_string_literal_size0(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 17, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -902,7 +1029,7 @@ define amdgpu_kernel void @test_print_string_literal_1ai8.zero(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 18, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -934,7 +1061,7 @@ define amdgpu_kernel void @test_print_string_literal_1ai8.undef(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 19, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -966,7 +1093,7 @@ define amdgpu_kernel void @test_print_string_literal_i8.zero(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 20, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -998,7 +1125,7 @@ define amdgpu_kernel void @test_print_string_literal_size2(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 21, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 27, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1030,7 +1157,7 @@ define amdgpu_kernel void @test_print_string_literal_size3(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 22, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 7195, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1062,7 +1189,7 @@ define amdgpu_kernel void @test_print_string_literal_size3_zero(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 23, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1094,7 +1221,7 @@ define amdgpu_kernel void @test_print_string_literal_size4(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 24, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 6513249, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1126,7 +1253,7 @@ define amdgpu_kernel void @test_print_string_literal_size4_nonull_term(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 25, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1158,7 +1285,7 @@ define amdgpu_kernel void @test_print_string_literal_size5(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 26, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1190,7 +1317,7 @@ define amdgpu_kernel void @test_print_string_literal_size6(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 27, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1225,7 +1352,7 @@ define amdgpu_kernel void @test_print_string_literal_size7(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 28, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1260,7 +1387,7 @@ define amdgpu_kernel void @test_print_string_literal_size8(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 29, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 35, 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
@@ -1295,7 +1422,7 @@ define amdgpu_kernel void @test_print_string_literal_size9(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 30, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 36, 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 1684104548, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1330,7 +1457,7 @@ define amdgpu_kernel void @test_print_string_literal_size16(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 31, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 909194801, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1371,7 +1498,7 @@ define amdgpu_kernel void @test_print_string_literal_size17(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 32, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 925972273, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1412,7 +1539,7 @@ define amdgpu_kernel void @test_print_string_literal_size20(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 33, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 1684234849, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1456,7 +1583,7 @@ define amdgpu_kernel void @test_print_string_literal_size32(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 34, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 40, 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
@@ -1509,7 +1636,7 @@ define amdgpu_kernel void @test_print_string_not_constant_global(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 35, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 41, 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
@@ -1541,7 +1668,7 @@ define amdgpu_kernel void @test_print_string_constant_interposable_global(i32 %n
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 36, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 42, 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
@@ -1573,7 +1700,7 @@ define amdgpu_kernel void @test_print_string_literal_v4i8(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 37, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 43, 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 67305985, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1605,7 +1732,7 @@ define amdgpu_kernel void @test_print_string_literal_v4i32(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 38, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 1, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1637,7 +1764,7 @@ define amdgpu_kernel void @test_print_string_literal_struct(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 39, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 66, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1669,7 +1796,7 @@ define amdgpu_kernel void @test_print_string_undef(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 40, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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
@@ -1701,7 +1828,7 @@ define amdgpu_kernel void @test_print_string_poison(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 41, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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
@@ -1733,7 +1860,7 @@ define amdgpu_kernel void @test_print_string_null(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 42, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1765,7 +1892,7 @@ define amdgpu_kernel void @test_print_string_inttoptr(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 43, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; 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 4144959, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1797,7 +1924,7 @@ define amdgpu_kernel void @test_print_string_float_neg0(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 44, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 50, 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1829,7 +1956,7 @@ define amdgpu_kernel void @test_print_string_float_0(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 45, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 51, 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1861,7 +1988,7 @@ define amdgpu_kernel void @test_print_string_ptr_null(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 46, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 52, 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1893,7 +2020,7 @@ define amdgpu_kernel void @test_print_string_ptr_undef(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 47, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 53, 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1925,7 +2052,7 @@ define amdgpu_kernel void @test_print_string_indexed(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 48, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 54, 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 1936875892, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1966,7 +2093,7 @@ define amdgpu_kernel void @test_print_string_indexed_oob(i32 %n) {
 ; GCN:       1:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 49, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 55, 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 -256, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 4
@@ -1982,6 +2109,65 @@ entry:
   ret void
 }
 
+@array.i16 = private unnamed_addr addrspace(4) constant [6 x i16] [i16 1, i16 2, i16 3, i16 4, i16 5, i16 6]
+
+; CHECK: error: <unknown>:0:0: in function test_format_array_i16 void (i32): printf format string must be a trivially resolved constant string global variable
+define amdgpu_kernel void @test_format_array_i16(i32 %n) {
+; R600-LABEL: @test_format_array_i16(
+; R600-NEXT:    [[STR:%.*]] = alloca [9 x i8], align 1, addrspace(5)
+; R600-NEXT:    [[CALL1:%.*]] = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) @array.i16, ptr addrspace(5) [[STR]], i32 [[N:%.*]])
+; R600-NEXT:    ret void
+;
+; GCN-LABEL: @test_format_array_i16(
+; GCN-NEXT:    [[STR:%.*]] = alloca [9 x i8], align 1, addrspace(5)
+; GCN-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 4)
+; GCN-NEXT:    br label [[DOTSPLIT:%.*]]
+; GCN:       .split:
+; GCN-NEXT:    [[TMP1:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
+; GCN-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]]
+; GCN:       2:
+; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
+; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
+; GCN-NEXT:    store i32 56, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
+; GCN-NEXT:    br label [[TMP3]]
+; GCN:       3:
+; GCN-NEXT:    ret void
+;
+  %str = alloca [9 x i8], align 1, addrspace(5)
+  %call1 = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) @array.i16, ptr addrspace(5) %str, i32 %n)
+  ret void
+}
+
+@struct = private unnamed_addr addrspace(4) constant { [6 x i8] } { [6 x i8] [i8 1, i8 2, i8 3, i8 4, i8 5, i8 6] }
+
+define amdgpu_kernel void @test_format_struct(i32 %n) {
+; R600-LABEL: @test_format_struct(
+; R600-NEXT:    [[STR:%.*]] = alloca [9 x i8], align 1, addrspace(5)
+; R600-NEXT:    [[CALL1:%.*]] = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) @struct, ptr addrspace(5) [[STR]], i32 [[N:%.*]])
+; R600-NEXT:    ret void
+;
+; GCN-LABEL: @test_format_struct(
+; GCN-NEXT:    [[STR:%.*]] = alloca [9 x i8], align 1, addrspace(5)
+; GCN-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 4)
+; GCN-NEXT:    br label [[DOTSPLIT:%.*]]
+; GCN:       .split:
+; GCN-NEXT:    [[TMP1:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
+; GCN-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]]
+; GCN:       2:
+; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
+; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
+; GCN-NEXT:    store i32 57, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    [[PRINTBUFFGEP:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
+; GCN-NEXT:    br label [[TMP3]]
+; GCN:       3:
+; GCN-NEXT:    ret void
+;
+  %str = alloca [9 x i8], align 1, addrspace(5)
+  %call1 = call i32 (ptr addrspace(4), ...) @printf(ptr addrspace(4) @struct, ptr addrspace(5) %str, i32 %n)
+  ret void
+}
+
 declare void @func(ptr)
 
 define void @not_call_operand_printf() {
@@ -2011,7 +2197,7 @@ define void @printf_printf(i32 %n) {
 ; GCN:       2:
 ; GCN-NEXT:    [[PRINTBUFFID:%.*]] = getelementptr i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 0
 ; GCN-NEXT:    [[PRINTBUFFIDCAST:%.*]] = bitcast ptr addrspace(1) [[PRINTBUFFID]] to ptr addrspace(1)
-; GCN-NEXT:    store i32 50, ptr addrspace(1) [[PRINTBUFFIDCAST]], align 4
+; GCN-NEXT:    store i32 58, 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 ptr @printf, ptr addrspace(1) [[PRINTBUFFPTRCAST]], align 8