// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
// REQUIRES: amdgpu-registered-target
// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -x hip -emit-llvm -mprintf-kind=buffered -fcuda-is-device \
// RUN:   -o - %s | FileCheck --enable-var-scope %s
// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -x hip -emit-llvm -ffp-exception-behavior=strict -mprintf-kind=buffered -fcuda-is-device \
// RUN:   -o - %s | FileCheck --enable-var-scope --check-prefix=CHECK_CONSTRAINED %s

#define __device__ __attribute__((device))
#define __shared__ __attribute__((shared))
#define __constant__ __attribute__((constant))

extern "C" __device__ int printf(const char *format, ...);

// CHECK-LABEL: define dso_local noundef i32 @_Z4foo1v
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[S:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[S_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[S]] to ptr
// CHECK-NEXT:    store ptr addrspacecast (ptr addrspace(4) @.str to ptr), ptr [[S_ASCAST]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ASCAST]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[S_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = icmp eq ptr [[TMP0]], null
// CHECK-NEXT:    br i1 [[TMP2]], label [[STRLEN_JOIN:%.*]], label [[STRLEN_WHILE:%.*]]
// CHECK:       strlen.while:
// CHECK-NEXT:    [[TMP3:%.*]] = phi ptr [ [[TMP0]], [[ENTRY:%.*]] ], [ [[TMP4:%.*]], [[STRLEN_WHILE]] ]
// CHECK-NEXT:    [[TMP4]] = getelementptr i8, ptr [[TMP3]], i64 1
// CHECK-NEXT:    [[TMP5:%.*]] = load i8, ptr [[TMP3]], align 1
// CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i8 [[TMP5]], 0
// CHECK-NEXT:    br i1 [[TMP6]], label [[STRLEN_WHILE_DONE:%.*]], label [[STRLEN_WHILE]]
// CHECK:       strlen.while.done:
// CHECK-NEXT:    [[TMP7:%.*]] = ptrtoaddr ptr [[TMP3]] to i64
// CHECK-NEXT:    [[TMP8:%.*]] = ptrtoaddr ptr [[TMP0]] to i64
// CHECK-NEXT:    [[TMP9:%.*]] = sub i64 [[TMP7]], [[TMP8]]
// CHECK-NEXT:    [[TMP10:%.*]] = add i64 [[TMP9]], 1
// CHECK-NEXT:    br label [[STRLEN_JOIN]]
// CHECK:       strlen.join:
// CHECK-NEXT:    [[TMP11:%.*]] = phi i64 [ [[TMP10]], [[STRLEN_WHILE_DONE]] ], [ 0, [[ENTRY]] ]
// CHECK-NEXT:    [[TMP12:%.*]] = add i64 [[TMP11]], 7
// CHECK-NEXT:    [[TMP13:%.*]] = and i64 [[TMP12]], 4294967288
// CHECK-NEXT:    [[TMP14:%.*]] = add i64 [[TMP13]], 52
// CHECK-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
// CHECK-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 [[TMP15]])
// CHECK-NEXT:    [[TMP16:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
// CHECK-NEXT:    br i1 [[TMP16]], label [[ARGPUSH_BLOCK:%.*]], label [[END_BLOCK:%.*]]
// CHECK:       end.block:
// CHECK-NEXT:    [[TMP17:%.*]] = xor i1 [[TMP16]], true
// CHECK-NEXT:    [[PRINTF_RESULT:%.*]] = sext i1 [[TMP17]] to i32
// CHECK-NEXT:    ret i32 [[PRINTF_RESULT]]
// CHECK:       argpush.block:
// CHECK-NEXT:    [[TMP18:%.*]] = shl i32 [[TMP15]], 2
// CHECK-NEXT:    [[TMP19:%.*]] = or i32 [[TMP18]], 2
// CHECK-NEXT:    store i32 [[TMP19]], ptr addrspace(1) [[PRINTF_ALLOC_FN]], align 4
// CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
// CHECK-NEXT:    store i64 1107004088646384690, ptr addrspace(1) [[TMP20]], align 8
// CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP20]], i32 8
// CHECK-NEXT:    store i64 8, ptr addrspace(1) [[TMP21]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 8
// CHECK-NEXT:    store double 3.141590e+00, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR1:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 8
// CHECK-NEXT:    store i64 8, ptr addrspace(1) [[PRINTBUFFNEXTPTR1]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR1]], i32 8
// CHECK-NEXT:    store i64 4, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR3:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 8
// CHECK-NEXT:    call void @llvm.memcpy.p1.p0.i64(ptr addrspace(1) align 1 [[PRINTBUFFNEXTPTR3]], ptr align 1 [[TMP0]], i64 [[TMP11]], i1 false)
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR3]], i64 [[TMP13]]
// CHECK-NEXT:    store ptr [[TMP1]], ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR5:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 8
// CHECK-NEXT:    br label [[END_BLOCK]]
//
// CHECK_CONSTRAINED-LABEL: define dso_local noundef i32 @_Z4foo1v
// CHECK_CONSTRAINED-NEXT:  entry:
// CHECK_CONSTRAINED-NEXT:    [[S:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK_CONSTRAINED-NEXT:    [[S_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[S]] to ptr
// CHECK_CONSTRAINED-NEXT:    store ptr addrspacecast (ptr addrspace(4) @.str to ptr), ptr [[S_ASCAST]], align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ASCAST]], align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[S_ASCAST]], align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP2:%.*]] = icmp eq ptr [[TMP0]], null
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP2]], label [[STRLEN_JOIN:%.*]], label [[STRLEN_WHILE:%.*]]
// CHECK_CONSTRAINED:       strlen.while:
// CHECK_CONSTRAINED-NEXT:    [[TMP3:%.*]] = phi ptr [ [[TMP0]], [[ENTRY:%.*]] ], [ [[TMP4:%.*]], [[STRLEN_WHILE]] ]
// CHECK_CONSTRAINED-NEXT:    [[TMP4]] = getelementptr i8, ptr [[TMP3]], i64 1
// CHECK_CONSTRAINED-NEXT:    [[TMP5:%.*]] = load i8, ptr [[TMP3]], align 1
// CHECK_CONSTRAINED-NEXT:    [[TMP6:%.*]] = icmp eq i8 [[TMP5]], 0
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP6]], label [[STRLEN_WHILE_DONE:%.*]], label [[STRLEN_WHILE]]
// CHECK_CONSTRAINED:       strlen.while.done:
// CHECK_CONSTRAINED-NEXT:    [[TMP7:%.*]] = ptrtoaddr ptr [[TMP3]] to i64
// CHECK_CONSTRAINED-NEXT:    [[TMP8:%.*]] = ptrtoaddr ptr [[TMP0]] to i64
// CHECK_CONSTRAINED-NEXT:    [[TMP9:%.*]] = sub i64 [[TMP7]], [[TMP8]]
// CHECK_CONSTRAINED-NEXT:    [[TMP10:%.*]] = add i64 [[TMP9]], 1
// CHECK_CONSTRAINED-NEXT:    br label [[STRLEN_JOIN]]
// CHECK_CONSTRAINED:       strlen.join:
// CHECK_CONSTRAINED-NEXT:    [[TMP11:%.*]] = phi i64 [ [[TMP10]], [[STRLEN_WHILE_DONE]] ], [ 0, [[ENTRY]] ]
// CHECK_CONSTRAINED-NEXT:    [[TMP12:%.*]] = add i64 [[TMP11]], 7
// CHECK_CONSTRAINED-NEXT:    [[TMP13:%.*]] = and i64 [[TMP12]], 4294967288
// CHECK_CONSTRAINED-NEXT:    [[TMP14:%.*]] = add i64 [[TMP13]], 52
// CHECK_CONSTRAINED-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
// CHECK_CONSTRAINED-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 [[TMP15]])
// CHECK_CONSTRAINED-NEXT:    [[TMP16:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP16]], label [[ARGPUSH_BLOCK:%.*]], label [[END_BLOCK:%.*]]
// CHECK_CONSTRAINED:       end.block:
// CHECK_CONSTRAINED-NEXT:    [[TMP17:%.*]] = xor i1 [[TMP16]], true
// CHECK_CONSTRAINED-NEXT:    [[PRINTF_RESULT:%.*]] = sext i1 [[TMP17]] to i32
// CHECK_CONSTRAINED-NEXT:    ret i32 [[PRINTF_RESULT]]
// CHECK_CONSTRAINED:       argpush.block:
// CHECK_CONSTRAINED-NEXT:    [[TMP18:%.*]] = shl i32 [[TMP15]], 2
// CHECK_CONSTRAINED-NEXT:    [[TMP19:%.*]] = or i32 [[TMP18]], 2
// CHECK_CONSTRAINED-NEXT:    store i32 [[TMP19]], ptr addrspace(1) [[PRINTF_ALLOC_FN]], align 4
// CHECK_CONSTRAINED-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
// CHECK_CONSTRAINED-NEXT:    store i64 1107004088646384690, ptr addrspace(1) [[TMP20]], align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP20]], i32 8
// CHECK_CONSTRAINED-NEXT:    store i64 8, ptr addrspace(1) [[TMP21]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 8
// CHECK_CONSTRAINED-NEXT:    store double 3.141590e+00, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR1:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 8
// CHECK_CONSTRAINED-NEXT:    store i64 8, ptr addrspace(1) [[PRINTBUFFNEXTPTR1]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR1]], i32 8
// CHECK_CONSTRAINED-NEXT:    store i64 4, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR3:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 8
// CHECK_CONSTRAINED-NEXT:    call void @llvm.memcpy.p1.p0.i64(ptr addrspace(1) align 1 [[PRINTBUFFNEXTPTR3]], ptr align 1 [[TMP0]], i64 [[TMP11]], i1 false)
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR3]], i64 [[TMP13]]
// CHECK_CONSTRAINED-NEXT:    store ptr [[TMP1]], ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR5:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 8
// CHECK_CONSTRAINED-NEXT:    br label [[END_BLOCK]]
//
__device__ int foo1() {
  const char *s = "hello world";
  return printf("%.*f %*.*s %p\n", 8, 3.14159, 8, 4, s, s);
}

__device__ char *dstr;
__device__ const
// CHECK-LABEL: define dso_local noundef i32 @_Z4foo2v
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LCVAL:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[LCVAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[LCVAL]] to ptr
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr addrspacecast (ptr addrspace(1) @dstr to ptr), align 8
// CHECK-NEXT:    [[TMP1:%.*]] = load ptr, ptr addrspacecast (ptr addrspace(1) @dstr to ptr), align 8
// CHECK-NEXT:    [[TMP2:%.*]] = icmp eq ptr [[TMP0]], null
// CHECK-NEXT:    br i1 [[TMP2]], label [[STRLEN_JOIN:%.*]], label [[STRLEN_WHILE:%.*]]
// CHECK:       strlen.while:
// CHECK-NEXT:    [[TMP3:%.*]] = phi ptr [ [[TMP0]], [[ENTRY:%.*]] ], [ [[TMP4:%.*]], [[STRLEN_WHILE]] ]
// CHECK-NEXT:    [[TMP4]] = getelementptr i8, ptr [[TMP3]], i64 1
// CHECK-NEXT:    [[TMP5:%.*]] = load i8, ptr [[TMP3]], align 1
// CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i8 [[TMP5]], 0
// CHECK-NEXT:    br i1 [[TMP6]], label [[STRLEN_WHILE_DONE:%.*]], label [[STRLEN_WHILE]]
// CHECK:       strlen.while.done:
// CHECK-NEXT:    [[TMP7:%.*]] = ptrtoaddr ptr [[TMP3]] to i64
// CHECK-NEXT:    [[TMP8:%.*]] = ptrtoaddr ptr [[TMP0]] to i64
// CHECK-NEXT:    [[TMP9:%.*]] = sub i64 [[TMP7]], [[TMP8]]
// CHECK-NEXT:    [[TMP10:%.*]] = add i64 [[TMP9]], 1
// CHECK-NEXT:    br label [[STRLEN_JOIN]]
// CHECK:       strlen.join:
// CHECK-NEXT:    [[TMP11:%.*]] = phi i64 [ [[TMP10]], [[STRLEN_WHILE_DONE]] ], [ 0, [[ENTRY]] ]
// CHECK-NEXT:    [[TMP12:%.*]] = add i64 [[TMP11]], 7
// CHECK-NEXT:    [[TMP13:%.*]] = and i64 [[TMP12]], 4294967288
// CHECK-NEXT:    [[TMP14:%.*]] = add i64 [[TMP13]], 36
// CHECK-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
// CHECK-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 [[TMP15]])
// CHECK-NEXT:    [[TMP16:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
// CHECK-NEXT:    br i1 [[TMP16]], label [[ARGPUSH_BLOCK:%.*]], label [[END_BLOCK:%.*]]
// CHECK:       end.block:
// CHECK-NEXT:    [[TMP17:%.*]] = xor i1 [[TMP16]], true
// CHECK-NEXT:    [[PRINTF_RESULT:%.*]] = sext i1 [[TMP17]] to i32
// CHECK-NEXT:    ret i32 [[PRINTF_RESULT]]
// CHECK:       argpush.block:
// CHECK-NEXT:    [[TMP18:%.*]] = shl i32 [[TMP15]], 2
// CHECK-NEXT:    [[TMP19:%.*]] = or i32 [[TMP18]], 2
// CHECK-NEXT:    store i32 [[TMP19]], ptr addrspace(1) [[PRINTF_ALLOC_FN]], align 4
// CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
// CHECK-NEXT:    store i64 7257695813269076350, ptr addrspace(1) [[TMP20]], align 8
// CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP20]], i32 8
// CHECK-NEXT:    call void @llvm.memcpy.p1.p0.i64(ptr addrspace(1) align 1 [[TMP21]], ptr align 1 [[TMP0]], i64 [[TMP11]], i1 false)
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i64 [[TMP13]]
// CHECK-NEXT:    store ptr [[TMP1]], ptr addrspace(1) [[PRINTBUFFNEXTPTR]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR1:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 8
// CHECK-NEXT:    store ptr addrspacecast (ptr addrspace(3) @_ZZ4foo2vE5shval to ptr), ptr addrspace(1) [[PRINTBUFFNEXTPTR1]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR1]], i32 8
// CHECK-NEXT:    store ptr [[LCVAL_ASCAST]], ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR3:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 8
// CHECK-NEXT:    br label [[END_BLOCK]]
//
// CHECK_CONSTRAINED-LABEL: define dso_local noundef i32 @_Z4foo2v
// CHECK_CONSTRAINED-NEXT:  entry:
// CHECK_CONSTRAINED-NEXT:    [[LCVAL:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK_CONSTRAINED-NEXT:    [[LCVAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[LCVAL]] to ptr
// CHECK_CONSTRAINED-NEXT:    [[TMP0:%.*]] = load ptr, ptr addrspacecast (ptr addrspace(1) @dstr to ptr), align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP1:%.*]] = load ptr, ptr addrspacecast (ptr addrspace(1) @dstr to ptr), align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP2:%.*]] = icmp eq ptr [[TMP0]], null
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP2]], label [[STRLEN_JOIN:%.*]], label [[STRLEN_WHILE:%.*]]
// CHECK_CONSTRAINED:       strlen.while:
// CHECK_CONSTRAINED-NEXT:    [[TMP3:%.*]] = phi ptr [ [[TMP0]], [[ENTRY:%.*]] ], [ [[TMP4:%.*]], [[STRLEN_WHILE]] ]
// CHECK_CONSTRAINED-NEXT:    [[TMP4]] = getelementptr i8, ptr [[TMP3]], i64 1
// CHECK_CONSTRAINED-NEXT:    [[TMP5:%.*]] = load i8, ptr [[TMP3]], align 1
// CHECK_CONSTRAINED-NEXT:    [[TMP6:%.*]] = icmp eq i8 [[TMP5]], 0
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP6]], label [[STRLEN_WHILE_DONE:%.*]], label [[STRLEN_WHILE]]
// CHECK_CONSTRAINED:       strlen.while.done:
// CHECK_CONSTRAINED-NEXT:    [[TMP7:%.*]] = ptrtoaddr ptr [[TMP3]] to i64
// CHECK_CONSTRAINED-NEXT:    [[TMP8:%.*]] = ptrtoaddr ptr [[TMP0]] to i64
// CHECK_CONSTRAINED-NEXT:    [[TMP9:%.*]] = sub i64 [[TMP7]], [[TMP8]]
// CHECK_CONSTRAINED-NEXT:    [[TMP10:%.*]] = add i64 [[TMP9]], 1
// CHECK_CONSTRAINED-NEXT:    br label [[STRLEN_JOIN]]
// CHECK_CONSTRAINED:       strlen.join:
// CHECK_CONSTRAINED-NEXT:    [[TMP11:%.*]] = phi i64 [ [[TMP10]], [[STRLEN_WHILE_DONE]] ], [ 0, [[ENTRY]] ]
// CHECK_CONSTRAINED-NEXT:    [[TMP12:%.*]] = add i64 [[TMP11]], 7
// CHECK_CONSTRAINED-NEXT:    [[TMP13:%.*]] = and i64 [[TMP12]], 4294967288
// CHECK_CONSTRAINED-NEXT:    [[TMP14:%.*]] = add i64 [[TMP13]], 36
// CHECK_CONSTRAINED-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
// CHECK_CONSTRAINED-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 [[TMP15]])
// CHECK_CONSTRAINED-NEXT:    [[TMP16:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP16]], label [[ARGPUSH_BLOCK:%.*]], label [[END_BLOCK:%.*]]
// CHECK_CONSTRAINED:       end.block:
// CHECK_CONSTRAINED-NEXT:    [[TMP17:%.*]] = xor i1 [[TMP16]], true
// CHECK_CONSTRAINED-NEXT:    [[PRINTF_RESULT:%.*]] = sext i1 [[TMP17]] to i32
// CHECK_CONSTRAINED-NEXT:    ret i32 [[PRINTF_RESULT]]
// CHECK_CONSTRAINED:       argpush.block:
// CHECK_CONSTRAINED-NEXT:    [[TMP18:%.*]] = shl i32 [[TMP15]], 2
// CHECK_CONSTRAINED-NEXT:    [[TMP19:%.*]] = or i32 [[TMP18]], 2
// CHECK_CONSTRAINED-NEXT:    store i32 [[TMP19]], ptr addrspace(1) [[PRINTF_ALLOC_FN]], align 4
// CHECK_CONSTRAINED-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
// CHECK_CONSTRAINED-NEXT:    store i64 7257695813269076350, ptr addrspace(1) [[TMP20]], align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP20]], i32 8
// CHECK_CONSTRAINED-NEXT:    call void @llvm.memcpy.p1.p0.i64(ptr addrspace(1) align 1 [[TMP21]], ptr align 1 [[TMP0]], i64 [[TMP11]], i1 false)
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i64 [[TMP13]]
// CHECK_CONSTRAINED-NEXT:    store ptr [[TMP1]], ptr addrspace(1) [[PRINTBUFFNEXTPTR]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR1:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 8
// CHECK_CONSTRAINED-NEXT:    store ptr addrspacecast (ptr addrspace(3) @_ZZ4foo2vE5shval to ptr), ptr addrspace(1) [[PRINTBUFFNEXTPTR1]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR2:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR1]], i32 8
// CHECK_CONSTRAINED-NEXT:    store ptr [[LCVAL_ASCAST]], ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR3:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR2]], i32 8
// CHECK_CONSTRAINED-NEXT:    br label [[END_BLOCK]]
//
__device__ int foo2() {
  __shared__ int shval;
  int lcval;
  return printf("%s %p %p %p\n", dstr, dstr, &shval, &lcval);
}

__device__ unsigned short g = 30;
__device__ unsigned long n = 30;

__device__ float f1 = 3.14f;
__device__ double f2 = 2.71828;
__device__ _Float16 f3 = 2.71;
__device__ __bf16 f4 = 3.142;
__device__ _BitInt(55) Int55 = 31;
__device__ _BitInt(44) Int44 = 312;
__device__ _BitInt(128) Int128 = 45637;

// CHECK-LABEL: define dso_local noundef i32 @_Z4foo3v
// CHECK-NEXT:  entry:
// CHECK-NEXT:    store i32 25, ptr addrspacecast (ptr addrspace(3) @_ZZ4foo3vE1s to ptr), align 4
// CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr addrspacecast (ptr addrspace(3) @_ZZ4foo3vE1s to ptr), align 4
// CHECK-NEXT:    [[TMP1:%.*]] = load i16, ptr addrspacecast (ptr addrspace(1) @g to ptr), align 2
// CHECK-NEXT:    [[CONV:%.*]] = zext i16 [[TMP1]] to i32
// CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr addrspacecast (ptr addrspace(1) @n to ptr), align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load float, ptr addrspacecast (ptr addrspace(1) @f1 to ptr), align 4
// CHECK-NEXT:    [[CONV1:%.*]] = fpext contract float [[TMP3]] to double
// CHECK-NEXT:    [[TMP4:%.*]] = load double, ptr addrspacecast (ptr addrspace(1) @f2 to ptr), align 8
// CHECK-NEXT:    [[TMP5:%.*]] = load half, ptr addrspacecast (ptr addrspace(1) @f3 to ptr), align 2
// CHECK-NEXT:    [[TMP6:%.*]] = load bfloat, ptr addrspacecast (ptr addrspace(1) @f4 to ptr), align 2
// CHECK-NEXT:    [[TMP7:%.*]] = load i64, ptr addrspacecast (ptr addrspace(1) @Int55 to ptr), align 8
// CHECK-NEXT:    [[LOADEDV:%.*]] = trunc i64 [[TMP7]] to i55
// CHECK-NEXT:    [[TMP8:%.*]] = load i64, ptr addrspacecast (ptr addrspace(1) @Int44 to ptr), align 8
// CHECK-NEXT:    [[LOADEDV2:%.*]] = trunc i64 [[TMP8]] to i44
// CHECK-NEXT:    [[TMP9:%.*]] = load i128, ptr addrspacecast (ptr addrspace(1) @Int128 to ptr), align 8
// CHECK-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 108)
// CHECK-NEXT:    [[TMP10:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
// CHECK-NEXT:    br i1 [[TMP10]], label [[ARGPUSH_BLOCK:%.*]], label [[END_BLOCK:%.*]]
// CHECK:       end.block:
// CHECK-NEXT:    [[TMP11:%.*]] = xor i1 [[TMP10]], true
// CHECK-NEXT:    [[PRINTF_RESULT:%.*]] = sext i1 [[TMP11]] to i32
// CHECK-NEXT:    ret i32 [[PRINTF_RESULT]]
// CHECK:       argpush.block:
// CHECK-NEXT:    store i32 434, ptr addrspace(1) [[PRINTF_ALLOC_FN]], align 4
// CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
// CHECK-NEXT:    store i64 7271852820361268873, ptr addrspace(1) [[TMP12]], align 8
// CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP12]], i32 8
// CHECK-NEXT:    [[TMP14:%.*]] = zext i32 [[TMP0]] to i64
// CHECK-NEXT:    store i64 [[TMP14]], ptr addrspace(1) [[TMP13]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP13]], i32 8
// CHECK-NEXT:    store ptr addrspacecast (ptr addrspace(3) @_ZZ4foo3vE1s to ptr), ptr addrspace(1) [[PRINTBUFFNEXTPTR]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR3:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 8
// CHECK-NEXT:    [[TMP15:%.*]] = zext i32 [[CONV]] to i64
// CHECK-NEXT:    store i64 [[TMP15]], ptr addrspace(1) [[PRINTBUFFNEXTPTR3]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR3]], i32 8
// CHECK-NEXT:    store i64 [[TMP2]], ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR5:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 8
// CHECK-NEXT:    store double [[CONV1]], ptr addrspace(1) [[PRINTBUFFNEXTPTR5]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR6:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR5]], i32 8
// CHECK-NEXT:    store double [[TMP4]], ptr addrspace(1) [[PRINTBUFFNEXTPTR6]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR7:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR6]], i32 8
// CHECK-NEXT:    [[TMP16:%.*]] = fpext half [[TMP5]] to double
// CHECK-NEXT:    store double [[TMP16]], ptr addrspace(1) [[PRINTBUFFNEXTPTR7]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR8:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR7]], i32 8
// CHECK-NEXT:    [[TMP17:%.*]] = fpext bfloat [[TMP6]] to double
// CHECK-NEXT:    store double [[TMP17]], ptr addrspace(1) [[PRINTBUFFNEXTPTR8]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR9:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR8]], i32 8
// CHECK-NEXT:    [[TMP18:%.*]] = zext i55 [[LOADEDV]] to i64
// CHECK-NEXT:    store i64 [[TMP18]], ptr addrspace(1) [[PRINTBUFFNEXTPTR9]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR10:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR9]], i32 8
// CHECK-NEXT:    [[TMP19:%.*]] = zext i44 [[LOADEDV2]] to i64
// CHECK-NEXT:    store i64 [[TMP19]], ptr addrspace(1) [[PRINTBUFFNEXTPTR10]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR11:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR10]], i32 8
// CHECK-NEXT:    store i128 [[TMP9]], ptr addrspace(1) [[PRINTBUFFNEXTPTR11]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR12:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR11]], i32 16
// CHECK-NEXT:    br label [[END_BLOCK]]
//
// CHECK_CONSTRAINED-LABEL: define dso_local noundef i32 @_Z4foo3v
// CHECK_CONSTRAINED-NEXT:  entry:
// CHECK_CONSTRAINED-NEXT:    store i32 25, ptr addrspacecast (ptr addrspace(3) @_ZZ4foo3vE1s to ptr), align 4
// CHECK_CONSTRAINED-NEXT:    [[TMP0:%.*]] = load i32, ptr addrspacecast (ptr addrspace(3) @_ZZ4foo3vE1s to ptr), align 4
// CHECK_CONSTRAINED-NEXT:    [[TMP1:%.*]] = load i16, ptr addrspacecast (ptr addrspace(1) @g to ptr), align 2
// CHECK_CONSTRAINED-NEXT:    [[CONV:%.*]] = zext i16 [[TMP1]] to i32
// CHECK_CONSTRAINED-NEXT:    [[TMP2:%.*]] = load i64, ptr addrspacecast (ptr addrspace(1) @n to ptr), align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP3:%.*]] = load float, ptr addrspacecast (ptr addrspace(1) @f1 to ptr), align 4
// CHECK_CONSTRAINED-NEXT:    [[CONV1:%.*]] = fpext contract float [[TMP3]] to double
// CHECK_CONSTRAINED-NEXT:    [[TMP4:%.*]] = load double, ptr addrspacecast (ptr addrspace(1) @f2 to ptr), align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP5:%.*]] = load half, ptr addrspacecast (ptr addrspace(1) @f3 to ptr), align 2
// CHECK_CONSTRAINED-NEXT:    [[TMP6:%.*]] = load bfloat, ptr addrspacecast (ptr addrspace(1) @f4 to ptr), align 2
// CHECK_CONSTRAINED-NEXT:    [[TMP7:%.*]] = load i64, ptr addrspacecast (ptr addrspace(1) @Int55 to ptr), align 8
// CHECK_CONSTRAINED-NEXT:    [[LOADEDV:%.*]] = trunc i64 [[TMP7]] to i55
// CHECK_CONSTRAINED-NEXT:    [[TMP8:%.*]] = load i64, ptr addrspacecast (ptr addrspace(1) @Int44 to ptr), align 8
// CHECK_CONSTRAINED-NEXT:    [[LOADEDV2:%.*]] = trunc i64 [[TMP8]] to i44
// CHECK_CONSTRAINED-NEXT:    [[TMP9:%.*]] = load i128, ptr addrspacecast (ptr addrspace(1) @Int128 to ptr), align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 108)
// CHECK_CONSTRAINED-NEXT:    [[TMP10:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP10]], label [[ARGPUSH_BLOCK:%.*]], label [[END_BLOCK:%.*]]
// CHECK_CONSTRAINED:       end.block:
// CHECK_CONSTRAINED-NEXT:    [[TMP11:%.*]] = xor i1 [[TMP10]], true
// CHECK_CONSTRAINED-NEXT:    [[PRINTF_RESULT:%.*]] = sext i1 [[TMP11]] to i32
// CHECK_CONSTRAINED-NEXT:    ret i32 [[PRINTF_RESULT]]
// CHECK_CONSTRAINED:       argpush.block:
// CHECK_CONSTRAINED-NEXT:    store i32 434, ptr addrspace(1) [[PRINTF_ALLOC_FN]], align 4
// CHECK_CONSTRAINED-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
// CHECK_CONSTRAINED-NEXT:    store i64 7271852820361268873, ptr addrspace(1) [[TMP12]], align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP12]], i32 8
// CHECK_CONSTRAINED-NEXT:    [[TMP14:%.*]] = zext i32 [[TMP0]] to i64
// CHECK_CONSTRAINED-NEXT:    store i64 [[TMP14]], ptr addrspace(1) [[TMP13]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP13]], i32 8
// CHECK_CONSTRAINED-NEXT:    store ptr addrspacecast (ptr addrspace(3) @_ZZ4foo3vE1s to ptr), ptr addrspace(1) [[PRINTBUFFNEXTPTR]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR3:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 8
// CHECK_CONSTRAINED-NEXT:    [[TMP15:%.*]] = zext i32 [[CONV]] to i64
// CHECK_CONSTRAINED-NEXT:    store i64 [[TMP15]], ptr addrspace(1) [[PRINTBUFFNEXTPTR3]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR4:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR3]], i32 8
// CHECK_CONSTRAINED-NEXT:    store i64 [[TMP2]], ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR5:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR4]], i32 8
// CHECK_CONSTRAINED-NEXT:    store double [[CONV1]], ptr addrspace(1) [[PRINTBUFFNEXTPTR5]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR6:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR5]], i32 8
// CHECK_CONSTRAINED-NEXT:    store double [[TMP4]], ptr addrspace(1) [[PRINTBUFFNEXTPTR6]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR7:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR6]], i32 8
// CHECK_CONSTRAINED-NEXT:    [[TMP16:%.*]] = fpext half [[TMP5]] to double
// CHECK_CONSTRAINED-NEXT:    store double [[TMP16]], ptr addrspace(1) [[PRINTBUFFNEXTPTR7]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR8:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR7]], i32 8
// CHECK_CONSTRAINED-NEXT:    [[TMP17:%.*]] = fpext bfloat [[TMP6]] to double
// CHECK_CONSTRAINED-NEXT:    store double [[TMP17]], ptr addrspace(1) [[PRINTBUFFNEXTPTR8]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR9:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR8]], i32 8
// CHECK_CONSTRAINED-NEXT:    [[TMP18:%.*]] = zext i55 [[LOADEDV]] to i64
// CHECK_CONSTRAINED-NEXT:    store i64 [[TMP18]], ptr addrspace(1) [[PRINTBUFFNEXTPTR9]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR10:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR9]], i32 8
// CHECK_CONSTRAINED-NEXT:    [[TMP19:%.*]] = zext i44 [[LOADEDV2]] to i64
// CHECK_CONSTRAINED-NEXT:    store i64 [[TMP19]], ptr addrspace(1) [[PRINTBUFFNEXTPTR10]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR11:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR10]], i32 8
// CHECK_CONSTRAINED-NEXT:    store i128 [[TMP9]], ptr addrspace(1) [[PRINTBUFFNEXTPTR11]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR12:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR11]], i32 16
// CHECK_CONSTRAINED-NEXT:    br label [[END_BLOCK]]
//
__device__ int foo3() {
  __shared__ int s;
  s = 25;
  return printf("Random values: %d,%p,%hd,%ld,%f,%f,%f,%f,%d,%d,%d\n",s, &s, g, n, f1, f2, f3, f4, Int55, Int44, Int128);
}

//A non trivial case,
// CHECK-LABEL: define dso_local noundef i32 @_Z4foo4v
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[S:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[S_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[S]] to ptr
// CHECK-NEXT:    store ptr addrspacecast (ptr addrspace(4) @.str.4 to ptr), ptr [[S_ASCAST]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ASCAST]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = icmp eq ptr [[TMP0]], null
// CHECK-NEXT:    br i1 [[TMP1]], label [[STRLEN_JOIN:%.*]], label [[STRLEN_WHILE:%.*]]
// CHECK:       strlen.while:
// CHECK-NEXT:    [[TMP2:%.*]] = phi ptr [ [[TMP0]], [[ENTRY:%.*]] ], [ [[TMP3:%.*]], [[STRLEN_WHILE]] ]
// CHECK-NEXT:    [[TMP3]] = getelementptr i8, ptr [[TMP2]], i64 1
// CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr [[TMP2]], align 1
// CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[TMP4]], 0
// CHECK-NEXT:    br i1 [[TMP5]], label [[STRLEN_WHILE_DONE:%.*]], label [[STRLEN_WHILE]]
// CHECK:       strlen.while.done:
// CHECK-NEXT:    [[TMP6:%.*]] = ptrtoaddr ptr [[TMP2]] to i64
// CHECK-NEXT:    [[TMP7:%.*]] = ptrtoaddr ptr [[TMP0]] to i64
// CHECK-NEXT:    [[TMP8:%.*]] = sub i64 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP9:%.*]] = add i64 [[TMP8]], 1
// CHECK-NEXT:    br label [[STRLEN_JOIN]]
// CHECK:       strlen.join:
// CHECK-NEXT:    [[TMP10:%.*]] = phi i64 [ [[TMP9]], [[STRLEN_WHILE_DONE]] ], [ 0, [[ENTRY]] ]
// CHECK-NEXT:    [[TMP11:%.*]] = add i64 [[TMP10]], 7
// CHECK-NEXT:    [[TMP12:%.*]] = and i64 [[TMP11]], 4294967288
// CHECK-NEXT:    [[TMP13:%.*]] = add i64 [[TMP12]], 12
// CHECK-NEXT:    [[TMP14:%.*]] = trunc i64 [[TMP13]] to i32
// CHECK-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 [[TMP14]])
// CHECK-NEXT:    [[TMP15:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
// CHECK-NEXT:    br i1 [[TMP15]], label [[ARGPUSH_BLOCK:%.*]], label [[END_BLOCK:%.*]]
// CHECK:       end.block:
// CHECK-NEXT:    [[TMP16:%.*]] = xor i1 [[TMP15]], true
// CHECK-NEXT:    [[PRINTF_RESULT:%.*]] = sext i1 [[TMP16]] to i32
// CHECK-NEXT:    ret i32 [[PRINTF_RESULT]]
// CHECK:       argpush.block:
// CHECK-NEXT:    [[TMP17:%.*]] = shl i32 [[TMP14]], 2
// CHECK-NEXT:    store i32 [[TMP17]], ptr addrspace(1) [[PRINTF_ALLOC_FN]], align 4
// CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
// CHECK-NEXT:    call void @llvm.memcpy.p1.p0.i64(ptr addrspace(1) align 1 [[TMP18]], ptr align 1 [[TMP0]], i64 [[TMP10]], i1 false)
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP18]], i64 [[TMP12]]
// CHECK-NEXT:    store i64 10, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], align 8
// CHECK-NEXT:    [[PRINTBUFFNEXTPTR1:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 8
// CHECK-NEXT:    br label [[END_BLOCK]]
//
// CHECK_CONSTRAINED-LABEL: define dso_local noundef i32 @_Z4foo4v
// CHECK_CONSTRAINED-NEXT:  entry:
// CHECK_CONSTRAINED-NEXT:    [[S:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK_CONSTRAINED-NEXT:    [[S_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[S]] to ptr
// CHECK_CONSTRAINED-NEXT:    store ptr addrspacecast (ptr addrspace(4) @.str.4 to ptr), ptr [[S_ASCAST]], align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[S_ASCAST]], align 8
// CHECK_CONSTRAINED-NEXT:    [[TMP1:%.*]] = icmp eq ptr [[TMP0]], null
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP1]], label [[STRLEN_JOIN:%.*]], label [[STRLEN_WHILE:%.*]]
// CHECK_CONSTRAINED:       strlen.while:
// CHECK_CONSTRAINED-NEXT:    [[TMP2:%.*]] = phi ptr [ [[TMP0]], [[ENTRY:%.*]] ], [ [[TMP3:%.*]], [[STRLEN_WHILE]] ]
// CHECK_CONSTRAINED-NEXT:    [[TMP3]] = getelementptr i8, ptr [[TMP2]], i64 1
// CHECK_CONSTRAINED-NEXT:    [[TMP4:%.*]] = load i8, ptr [[TMP2]], align 1
// CHECK_CONSTRAINED-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[TMP4]], 0
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP5]], label [[STRLEN_WHILE_DONE:%.*]], label [[STRLEN_WHILE]]
// CHECK_CONSTRAINED:       strlen.while.done:
// CHECK_CONSTRAINED-NEXT:    [[TMP6:%.*]] = ptrtoaddr ptr [[TMP2]] to i64
// CHECK_CONSTRAINED-NEXT:    [[TMP7:%.*]] = ptrtoaddr ptr [[TMP0]] to i64
// CHECK_CONSTRAINED-NEXT:    [[TMP8:%.*]] = sub i64 [[TMP6]], [[TMP7]]
// CHECK_CONSTRAINED-NEXT:    [[TMP9:%.*]] = add i64 [[TMP8]], 1
// CHECK_CONSTRAINED-NEXT:    br label [[STRLEN_JOIN]]
// CHECK_CONSTRAINED:       strlen.join:
// CHECK_CONSTRAINED-NEXT:    [[TMP10:%.*]] = phi i64 [ [[TMP9]], [[STRLEN_WHILE_DONE]] ], [ 0, [[ENTRY]] ]
// CHECK_CONSTRAINED-NEXT:    [[TMP11:%.*]] = add i64 [[TMP10]], 7
// CHECK_CONSTRAINED-NEXT:    [[TMP12:%.*]] = and i64 [[TMP11]], 4294967288
// CHECK_CONSTRAINED-NEXT:    [[TMP13:%.*]] = add i64 [[TMP12]], 12
// CHECK_CONSTRAINED-NEXT:    [[TMP14:%.*]] = trunc i64 [[TMP13]] to i32
// CHECK_CONSTRAINED-NEXT:    [[PRINTF_ALLOC_FN:%.*]] = call ptr addrspace(1) @__printf_alloc(i32 [[TMP14]])
// CHECK_CONSTRAINED-NEXT:    [[TMP15:%.*]] = icmp ne ptr addrspace(1) [[PRINTF_ALLOC_FN]], null
// CHECK_CONSTRAINED-NEXT:    br i1 [[TMP15]], label [[ARGPUSH_BLOCK:%.*]], label [[END_BLOCK:%.*]]
// CHECK_CONSTRAINED:       end.block:
// CHECK_CONSTRAINED-NEXT:    [[TMP16:%.*]] = xor i1 [[TMP15]], true
// CHECK_CONSTRAINED-NEXT:    [[PRINTF_RESULT:%.*]] = sext i1 [[TMP16]] to i32
// CHECK_CONSTRAINED-NEXT:    ret i32 [[PRINTF_RESULT]]
// CHECK_CONSTRAINED:       argpush.block:
// CHECK_CONSTRAINED-NEXT:    [[TMP17:%.*]] = shl i32 [[TMP14]], 2
// CHECK_CONSTRAINED-NEXT:    store i32 [[TMP17]], ptr addrspace(1) [[PRINTF_ALLOC_FN]], align 4
// CHECK_CONSTRAINED-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTF_ALLOC_FN]], i32 4
// CHECK_CONSTRAINED-NEXT:    call void @llvm.memcpy.p1.p0.i64(ptr addrspace(1) align 1 [[TMP18]], ptr align 1 [[TMP0]], i64 [[TMP10]], i1 false)
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP18]], i64 [[TMP12]]
// CHECK_CONSTRAINED-NEXT:    store i64 10, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], align 8
// CHECK_CONSTRAINED-NEXT:    [[PRINTBUFFNEXTPTR1:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[PRINTBUFFNEXTPTR]], i32 8
// CHECK_CONSTRAINED-NEXT:    br label [[END_BLOCK]]
//
__device__ int foo4() {
  const char* s = "format str%d";
  return printf(s, 10);
}