; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals
; RUN: opt -mtriple=amdgcn-amd-amdhsa -S -o - -passes=amdgpu-lower-kernel-arguments %s | FileCheck -check-prefixes=GCN,HSA %s
; RUN: opt -mtriple=amdgcn-- -S -o - -passes=amdgpu-lower-kernel-arguments %s | FileCheck -check-prefixes=GCN,MESA %s

target datalayout = "A5"

declare void @llvm.fake.use(...)

define amdgpu_kernel void @kern_noargs() {
; GCN-LABEL: @kern_noargs(
; GCN-NEXT:    ret void
;
  ret void
}

define amdgpu_kernel void @kern_i8(i8 %arg) #0 {
; HSA-LABEL: @kern_i8(
; HSA-NEXT:    [[KERN_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1:![0-9]+]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    store i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i8(
; MESA-NEXT:    [[KERN_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1:![0-9]+]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    store i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store i8 %arg, ptr addrspace(1) undef, align 1
  ret void
}

define amdgpu_kernel void @kern_i16(i16 %arg) #0 {
; HSA-LABEL: @kern_i16(
; HSA-NEXT:    [[KERN_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    store i16 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i16(
; MESA-NEXT:    [[KERN_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    store i16 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store i16 %arg, ptr addrspace(1) undef, align 1
  ret void
}

define amdgpu_kernel void @kern_f16(half %arg) #0 {
; HSA-LABEL: @kern_f16(
; HSA-NEXT:    [[KERN_F16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_F16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    [[ARG_LOAD:%.*]] = bitcast i16 [[TMP2]] to half
; HSA-NEXT:    store half [[ARG_LOAD]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_f16(
; MESA-NEXT:    [[KERN_F16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_F16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    [[ARG_LOAD:%.*]] = bitcast i16 [[TMP2]] to half
; MESA-NEXT:    store half [[ARG_LOAD]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store half %arg, ptr addrspace(1) undef, align 1
  ret void
}

define amdgpu_kernel void @kern_zeroext_i8(i8 zeroext %arg) #0 {
; HSA-LABEL: @kern_zeroext_i8(
; HSA-NEXT:    [[KERN_ZEROEXT_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_ZEROEXT_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    store i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_zeroext_i8(
; MESA-NEXT:    [[KERN_ZEROEXT_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_ZEROEXT_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    store i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store i8 %arg, ptr addrspace(1) undef, align 1
  ret void
}

define amdgpu_kernel void @kern_zeroext_i16(i16 zeroext %arg) #0 {
; HSA-LABEL: @kern_zeroext_i16(
; HSA-NEXT:    [[KERN_ZEROEXT_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_ZEROEXT_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    store i16 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_zeroext_i16(
; MESA-NEXT:    [[KERN_ZEROEXT_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_ZEROEXT_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    store i16 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store i16 %arg, ptr addrspace(1) undef, align 1
  ret void
}

define amdgpu_kernel void @kern_signext_i8(i8 signext %arg) #0 {
; HSA-LABEL: @kern_signext_i8(
; HSA-NEXT:    [[KERN_SIGNEXT_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_SIGNEXT_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    store i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_signext_i8(
; MESA-NEXT:    [[KERN_SIGNEXT_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_SIGNEXT_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    store i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store i8 %arg, ptr addrspace(1) undef, align 1
  ret void
}

define amdgpu_kernel void @kern_signext_i16(i16 signext %arg) #0 {
; HSA-LABEL: @kern_signext_i16(
; HSA-NEXT:    [[KERN_SIGNEXT_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_SIGNEXT_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    store i16 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_signext_i16(
; MESA-NEXT:    [[KERN_SIGNEXT_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_SIGNEXT_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    store i16 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store i16 %arg, ptr addrspace(1) undef, align 1
  ret void
}

define amdgpu_kernel void @kern_i8_i8(i8 %arg0, i8 %arg1) {
; HSA-LABEL: @kern_i8_i8(
; HSA-NEXT:    [[KERN_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i8_i8(
; MESA-NEXT:    [[KERN_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, ptr addrspace(1) undef, align 1
  store volatile i8 %arg1, ptr addrspace(1) undef, align 1
  ret void
}

define amdgpu_kernel void @kern_v3i8(<3 x i8> %arg) {
; HSA-LABEL: @kern_v3i8(
; HSA-NEXT:    [[KERN_V3I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V3I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i24
; HSA-NEXT:    [[ARG_LOAD:%.*]] = bitcast i24 [[TMP2]] to <3 x i8>
; HSA-NEXT:    store <3 x i8> [[ARG_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v3i8(
; MESA-NEXT:    [[KERN_V3I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V3I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i24
; MESA-NEXT:    [[ARG_LOAD:%.*]] = bitcast i24 [[TMP2]] to <3 x i8>
; MESA-NEXT:    store <3 x i8> [[ARG_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store <3 x i8> %arg, ptr addrspace(1) undef, align 4
  ret void
}

define amdgpu_kernel void @kern_i24(i24 %arg0) {
; HSA-LABEL: @kern_i24(
; HSA-NEXT:    [[KERN_I24_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I24_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i24
; HSA-NEXT:    store i24 [[TMP2]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i24(
; MESA-NEXT:    [[KERN_I24_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I24_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i24
; MESA-NEXT:    store i24 [[TMP2]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store i24 %arg0, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_i32(i32 %arg0) {
; HSA-LABEL: @kern_i32(
; HSA-NEXT:    [[KERN_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store i32 [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i32(
; MESA-NEXT:    [[KERN_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store i32 [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store i32 %arg0, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_range_noundef_i32(i32 noundef range(i32 0, 8) %arg0) {
; HSA-LABEL: @kern_range_noundef_i32(
; HSA-NEXT:    [[KERN_RANGE_NOUNDEF_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_RANGE_NOUNDEF_I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !range [[RNG2:![0-9]+]], !invariant.load [[META1]], !noundef [[META1]]
; HSA-NEXT:    call void (...) @llvm.fake.use(i32 [[ARG0_LOAD]])
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_range_noundef_i32(
; MESA-NEXT:    [[KERN_RANGE_NOUNDEF_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_RANGE_NOUNDEF_I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !range [[RNG2:![0-9]+]], !invariant.load [[META1]], !noundef [[META1]]
; MESA-NEXT:    call void (...) @llvm.fake.use(i32 [[ARG0_LOAD]])
; MESA-NEXT:    ret void
;
  call void (...) @llvm.fake.use(i32 %arg0)
  ret void
}

define amdgpu_kernel void @kern_f32(float %arg0) {
; HSA-LABEL: @kern_f32(
; HSA-NEXT:    [[KERN_F32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_F32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load float, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store float [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_f32(
; MESA-NEXT:    [[KERN_F32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_F32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load float, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store float [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store float %arg0, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_v3i32(<3 x i32> %arg0) {
; HSA-LABEL: @kern_v3i32(
; HSA-NEXT:    [[KERN_V3I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V3I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
; HSA-NEXT:    store <3 x i32> [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v3i32(
; MESA-NEXT:    [[KERN_V3I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V3I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
; MESA-NEXT:    store <3 x i32> [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store <3 x i32> %arg0, ptr addrspace(1) undef, align 4
  ret void
}

define amdgpu_kernel void @kern_v8i32(<8 x i32> %arg) #0 {
; HSA-LABEL: @kern_v8i32(
; HSA-NEXT:    [[KERN_V8I32_KERNARG_SEGMENT:%.*]] = call nonnull align 32 dereferenceable(288) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V8I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_LOAD:%.*]] = load <8 x i32>, ptr addrspace(4) [[ARG_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store <8 x i32> [[ARG_LOAD]], ptr addrspace(1) undef, align 32
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v8i32(
; MESA-NEXT:    [[KERN_V8I32_KERNARG_SEGMENT:%.*]] = call nonnull align 32 dereferenceable(288) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V8I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_LOAD:%.*]] = load <8 x i32>, ptr addrspace(4) [[ARG_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store <8 x i32> [[ARG_LOAD]], ptr addrspace(1) undef, align 32
; MESA-NEXT:    ret void
;
  store <8 x i32> %arg, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_v8i64(<8 x i64> %arg) #0 {
; HSA-LABEL: @kern_v8i64(
; HSA-NEXT:    [[KERN_V8I64_KERNARG_SEGMENT:%.*]] = call nonnull align 64 dereferenceable(320) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V8I64_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_LOAD:%.*]] = load <8 x i64>, ptr addrspace(4) [[ARG_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store <8 x i64> [[ARG_LOAD]], ptr addrspace(1) undef, align 64
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v8i64(
; MESA-NEXT:    [[KERN_V8I64_KERNARG_SEGMENT:%.*]] = call nonnull align 64 dereferenceable(320) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V8I64_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_LOAD:%.*]] = load <8 x i64>, ptr addrspace(4) [[ARG_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store <8 x i64> [[ARG_LOAD]], ptr addrspace(1) undef, align 64
; MESA-NEXT:    ret void
;
  store <8 x i64> %arg, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_v16i64(<16 x i64> %arg) #0 {
; HSA-LABEL: @kern_v16i64(
; HSA-NEXT:    [[KERN_V16I64_KERNARG_SEGMENT:%.*]] = call nonnull align 128 dereferenceable(384) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V16I64_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_LOAD:%.*]] = load <16 x i64>, ptr addrspace(4) [[ARG_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store <16 x i64> [[ARG_LOAD]], ptr addrspace(1) undef, align 128
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v16i64(
; MESA-NEXT:    [[KERN_V16I64_KERNARG_SEGMENT:%.*]] = call nonnull align 128 dereferenceable(384) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_V16I64_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_LOAD:%.*]] = load <16 x i64>, ptr addrspace(4) [[ARG_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store <16 x i64> [[ARG_LOAD]], ptr addrspace(1) undef, align 128
; MESA-NEXT:    ret void
;
  store <16 x i64> %arg, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_i32_v3i32(i32 %arg0, <3 x i32> %arg1) {
; HSA-LABEL: @kern_i32_v3i32(
; HSA-NEXT:    [[KERN_I32_V3I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(288) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I32_V3I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I32_V3I32_KERNARG_SEGMENT]], i64 16
; HSA-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr addrspace(4) [[ARG1_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
; HSA-NEXT:    store i32 [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    store <3 x i32> [[ARG1_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i32_v3i32(
; MESA-NEXT:    [[KERN_I32_V3I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(288) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I32_V3I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_I32_V3I32_KERNARG_SEGMENT]], i64 52
; MESA-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr addrspace(4) [[ARG1_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
; MESA-NEXT:    store i32 [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    store <3 x i32> [[ARG1_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store i32 %arg0, ptr addrspace(1) undef
  store <3 x i32> %arg1, ptr addrspace(1) undef, align 4
  ret void
}

%struct.a = type { i32, i8, [4 x i8] }
%struct.b.packed = type { i8, i32, [3 x i16], <2 x double> }

define amdgpu_kernel void @kern_struct_a(%struct.a %arg0) {
; HSA-LABEL: @kern_struct_a(
; HSA-NEXT:    [[KERN_STRUCT_A_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_STRUCT_A_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load [[STRUCT_A:%.*]], ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store [[STRUCT_A]] [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_struct_a(
; MESA-NEXT:    [[KERN_STRUCT_A_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(268) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_STRUCT_A_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load [[STRUCT_A:%.*]], ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store [[STRUCT_A]] [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store %struct.a %arg0, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_struct_b_packed(%struct.b.packed %arg0) #0 {
; HSA-LABEL: @kern_struct_b_packed(
; HSA-NEXT:    [[KERN_STRUCT_B_PACKED_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(288) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_STRUCT_B_PACKED_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load [[STRUCT_B_PACKED:%.*]], ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store [[STRUCT_B_PACKED]] [[ARG0_LOAD]], ptr addrspace(1) undef, align 16
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_struct_b_packed(
; MESA-NEXT:    [[KERN_STRUCT_B_PACKED_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(288) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_STRUCT_B_PACKED_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load [[STRUCT_B_PACKED:%.*]], ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store [[STRUCT_B_PACKED]] [[ARG0_LOAD]], ptr addrspace(1) undef, align 16
; MESA-NEXT:    ret void
;
  store %struct.b.packed %arg0, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_implicit_arg_num_bytes(i32 %arg0) #1 {
; HSA-LABEL: @kern_implicit_arg_num_bytes(
; HSA-NEXT:    [[KERN_IMPLICIT_ARG_NUM_BYTES_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(48) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_IMPLICIT_ARG_NUM_BYTES_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store i32 [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_implicit_arg_num_bytes(
; MESA-NEXT:    [[KERN_IMPLICIT_ARG_NUM_BYTES_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_IMPLICIT_ARG_NUM_BYTES_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store i32 [[ARG0_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store i32 %arg0, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kernel_implicitarg_no_struct_align(<16 x i32>, i32 %arg1) #1 {
; HSA-LABEL: @kernel_implicitarg_no_struct_align(
; HSA-NEXT:    [[KERNEL_IMPLICITARG_NO_STRUCT_ALIGN_KERNARG_SEGMENT:%.*]] = call nonnull align 64 dereferenceable(112) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERNEL_IMPLICITARG_NO_STRUCT_ALIGN_KERNARG_SEGMENT]], i64 64
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store i32 [[ARG1_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kernel_implicitarg_no_struct_align(
; MESA-NEXT:    [[KERNEL_IMPLICITARG_NO_STRUCT_ALIGN_KERNARG_SEGMENT:%.*]] = call nonnull align 64 dereferenceable(108) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERNEL_IMPLICITARG_NO_STRUCT_ALIGN_KERNARG_SEGMENT]], i64 100
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store i32 [[ARG1_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store i32 %arg1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_lds_ptr(ptr addrspace(3) %lds) #0 {
; HSA-LABEL: @kern_lds_ptr(
; HSA-NEXT:    [[KERN_LDS_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[LDS_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_LDS_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[LDS_LOAD:%.*]] = load ptr addrspace(3), ptr addrspace(4) [[LDS_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store i32 0, ptr addrspace(3) [[LDS_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_lds_ptr(
; MESA-NEXT:    [[KERN_LDS_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[LDS_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_LDS_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[LDS_LOAD:%.*]] = load ptr addrspace(3), ptr addrspace(4) [[LDS_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store i32 0, ptr addrspace(3) [[LDS_LOAD]], align 4
; MESA-NEXT:    ret void
;
  store i32 0, ptr addrspace(3) %lds, align 4
  ret void
}

define amdgpu_kernel void @kern_lds_ptr_si(ptr addrspace(3) %lds) #2 {
; GCN-LABEL: @kern_lds_ptr_si(
; GCN-NEXT:    [[KERN_LDS_PTR_SI_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; GCN-NEXT:    store i32 0, ptr addrspace(3) [[LDS:%.*]], align 4
; GCN-NEXT:    ret void
;
  store i32 0, ptr addrspace(3) %lds, align 4
  ret void
}

define amdgpu_kernel void @kern_realign_i8_i8(i8 %arg0, i8 %arg1) #0 {
; HSA-LABEL: @kern_realign_i8_i8(
; HSA-NEXT:    [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_i8(
; MESA-NEXT:    [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, ptr addrspace(1) undef
  store volatile i8 %arg1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i8_i8_i8(i8 %arg0, i8 %arg1, i8 %arg2) #0 {
; HSA-LABEL: @kern_realign_i8_i8_i8(
; HSA-NEXT:    [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; HSA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP8]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_i8_i8(
; MESA-NEXT:    [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; MESA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP8]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, ptr addrspace(1) undef
  store volatile i8 %arg1, ptr addrspace(1) undef
  store volatile i8 %arg2, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i8_i8_i8_i8(i8 %arg0, i8 %arg1, i8 %arg2, i8 %arg3) #0 {
; HSA-LABEL: @kern_realign_i8_i8_i8_i8(
; HSA-NEXT:    [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; HSA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i8
; HSA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP9:%.*]] = load i32, ptr addrspace(4) [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; HSA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP8]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP11]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_i8_i8_i8(
; MESA-NEXT:    [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; MESA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i8
; MESA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP9:%.*]] = load i32, ptr addrspace(4) [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; MESA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP8]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP11]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, ptr addrspace(1) undef
  store volatile i8 %arg1, ptr addrspace(1) undef
  store volatile i8 %arg2, ptr addrspace(1) undef
  store volatile i8 %arg3, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i8_v3i8(i8 %arg0, <3 x i8> %arg1) #0 {
; HSA-LABEL: @kern_realign_i8_v3i8(
; HSA-NEXT:    [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = trunc i32 [[TMP3]] to i24
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = bitcast i24 [[TMP4]] to <3 x i8>
; HSA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile <3 x i8> [[ARG1_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_v3i8(
; MESA-NEXT:    [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 8, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = trunc i32 [[TMP3]] to i24
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = bitcast i24 [[TMP4]] to <3 x i8>
; MESA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile <3 x i8> [[ARG1_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, ptr addrspace(1) undef
  store volatile <3 x i8> %arg1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i8_i16(i8 %arg0, i16 %arg1) #0 {
; HSA-LABEL: @kern_realign_i8_i16(
; HSA-NEXT:    [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; HSA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i16 [[TMP5]], ptr addrspace(1) undef, align 2
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_i16(
; MESA-NEXT:    [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; MESA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i16 [[TMP5]], ptr addrspace(1) undef, align 2
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, ptr addrspace(1) undef
  store volatile i16 %arg1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i1_i1(i1 %arg0, i1 %arg1) #0 {
; HSA-LABEL: @kern_realign_i1_i1(
; HSA-NEXT:    [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i1
; HSA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i1 [[TMP5]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i1_i1(
; MESA-NEXT:    [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i1
; MESA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i1 [[TMP5]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, ptr addrspace(1) undef
  store volatile i1 %arg1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i1_i1_i1(i1 %arg0, i1 %arg1, i1 %arg2) #0 {
; HSA-LABEL: @kern_realign_i1_i1_i1(
; HSA-NEXT:    [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i1
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; HSA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i1
; HSA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i1 [[TMP5]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i1 [[TMP8]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i1_i1_i1(
; MESA-NEXT:    [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i1
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; MESA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i1
; MESA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i1 [[TMP5]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i1 [[TMP8]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, ptr addrspace(1) undef
  store volatile i1 %arg1, ptr addrspace(1) undef
  store volatile i1 %arg2, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i1_i1_i1_i1(i1 %arg0, i1 %arg1, i1 %arg2, i1 %arg3) #0 {
; HSA-LABEL: @kern_realign_i1_i1_i1_i1(
; HSA-NEXT:    [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i1
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; HSA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i1
; HSA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP9:%.*]] = load i32, ptr addrspace(4) [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; HSA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i1
; HSA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i1 [[TMP5]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i1 [[TMP8]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i1 [[TMP11]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i1_i1_i1_i1(
; MESA-NEXT:    [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i1
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; MESA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i1
; MESA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP9:%.*]] = load i32, ptr addrspace(4) [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; MESA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i1
; MESA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i1 [[TMP5]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i1 [[TMP8]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i1 [[TMP11]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, ptr addrspace(1) undef
  store volatile i1 %arg1, ptr addrspace(1) undef
  store volatile i1 %arg2, ptr addrspace(1) undef
  store volatile i1 %arg3, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i1_v3i1(i1 %arg0, <3 x i1> %arg1) #0 {
; HSA-LABEL: @kern_realign_i1_v3i1(
; HSA-NEXT:    [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i3
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = bitcast i3 [[TMP5]] to <3 x i1>
; HSA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile <3 x i1> [[ARG1_LOAD]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i1_v3i1(
; MESA-NEXT:    [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i3
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = bitcast i3 [[TMP5]] to <3 x i1>
; MESA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile <3 x i1> [[ARG1_LOAD]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, ptr addrspace(1) undef
  store volatile <3 x i1> %arg1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i1_i16(i1 %arg0, i16 %arg1) #0 {
; HSA-LABEL: @kern_realign_i1_i16(
; HSA-NEXT:    [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; HSA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i16 [[TMP5]], ptr addrspace(1) undef, align 2
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i1_i16(
; MESA-NEXT:    [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; MESA-NEXT:    store volatile i1 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i16 [[TMP5]], ptr addrspace(1) undef, align 2
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, ptr addrspace(1) undef
  store volatile i16 %arg1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_i8_i8_i8_i8_i8_i8_i8_i8(i8 %arg0, i8 %arg1, i8 %arg2, i8 %arg3, i8 %arg4, i8 %arg5, i8 %arg6, i8 %arg7) #0 {
; HSA-LABEL: @kern_realign_i8_i8_i8_i8_i8_i8_i8_i8(
; HSA-NEXT:    [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; HSA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i8
; HSA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP9:%.*]] = load i32, ptr addrspace(4) [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; HSA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i8
; HSA-NEXT:    [[ARG5_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[TMP12:%.*]] = load i32, ptr addrspace(4) [[ARG5_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP13:%.*]] = lshr i32 [[TMP12]], 8
; HSA-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i8
; HSA-NEXT:    [[ARG6_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[TMP15:%.*]] = load i32, ptr addrspace(4) [[ARG6_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP16:%.*]] = lshr i32 [[TMP15]], 16
; HSA-NEXT:    [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8
; HSA-NEXT:    [[ARG7_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[TMP18:%.*]] = load i32, ptr addrspace(4) [[ARG7_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP19:%.*]] = lshr i32 [[TMP18]], 24
; HSA-NEXT:    [[TMP20:%.*]] = trunc i32 [[TMP19]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP8]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP11]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP14]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP17]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP20]], ptr addrspace(1) undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_i8_i8_i8_i8_i8_i8_i8(
; MESA-NEXT:    [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(4) [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; MESA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i8
; MESA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP9:%.*]] = load i32, ptr addrspace(4) [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; MESA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i8
; MESA-NEXT:    [[ARG5_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[TMP12:%.*]] = load i32, ptr addrspace(4) [[ARG5_KERNARG_OFFSET_ALIGN_DOWN]], align 8, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP13:%.*]] = lshr i32 [[TMP12]], 8
; MESA-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i8
; MESA-NEXT:    [[ARG6_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[TMP15:%.*]] = load i32, ptr addrspace(4) [[ARG6_KERNARG_OFFSET_ALIGN_DOWN]], align 8, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP16:%.*]] = lshr i32 [[TMP15]], 16
; MESA-NEXT:    [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8
; MESA-NEXT:    [[ARG7_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[TMP18:%.*]] = load i32, ptr addrspace(4) [[ARG7_KERNARG_OFFSET_ALIGN_DOWN]], align 8, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP19:%.*]] = lshr i32 [[TMP18]], 24
; MESA-NEXT:    [[TMP20:%.*]] = trunc i32 [[TMP19]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP5]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP8]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP11]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP14]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP17]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP20]], ptr addrspace(1) undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, ptr addrspace(1) undef
  store volatile i8 %arg1, ptr addrspace(1) undef
  store volatile i8 %arg2, ptr addrspace(1) undef
  store volatile i8 %arg3, ptr addrspace(1) undef
  store volatile i8 %arg5, ptr addrspace(1) undef
  store volatile i8 %arg6, ptr addrspace(1) undef
  store volatile i8 %arg7, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_realign_f16_f16(half %arg0, half %arg1) #0 {
; HSA-LABEL: @kern_realign_f16_f16(
; HSA-NEXT:    [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = bitcast i16 [[TMP2]] to half
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = bitcast i16 [[TMP5]] to half
; HSA-NEXT:    store volatile half [[ARG0_LOAD]], ptr addrspace(1) undef, align 2
; HSA-NEXT:    store volatile half [[ARG1_LOAD]], ptr addrspace(1) undef, align 2
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_f16_f16(
; MESA-NEXT:    [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = bitcast i16 [[TMP2]] to half
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = bitcast i16 [[TMP5]] to half
; MESA-NEXT:    store volatile half [[ARG0_LOAD]], ptr addrspace(1) undef, align 2
; MESA-NEXT:    store volatile half [[ARG1_LOAD]], ptr addrspace(1) undef, align 2
; MESA-NEXT:    ret void
;
  store volatile half %arg0, ptr addrspace(1) undef
  store volatile half %arg1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_global_ptr(ptr addrspace(1) %ptr) #0 {
; HSA-LABEL: @kern_global_ptr(
; HSA-NEXT:    [[KERN_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_GLOBAL_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_global_ptr(
; MESA-NEXT:    [[KERN_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_GLOBAL_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; MESA-NEXT:    ret void
;
  store volatile ptr addrspace(1) %ptr, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_global_ptr_dereferencable(ptr addrspace(1) dereferenceable(42) %ptr) #0 {
; HSA-LABEL: @kern_global_ptr_dereferencable(
; HSA-NEXT:    [[KERN_GLOBAL_PTR_DEREFERENCABLE_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_GLOBAL_PTR_DEREFERENCABLE_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !dereferenceable [[META3:![0-9]+]]
; HSA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_global_ptr_dereferencable(
; MESA-NEXT:    [[KERN_GLOBAL_PTR_DEREFERENCABLE_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_GLOBAL_PTR_DEREFERENCABLE_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !dereferenceable [[META3:![0-9]+]]
; MESA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; MESA-NEXT:    ret void
;
  store volatile ptr addrspace(1) %ptr, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_global_ptr_dereferencable_or_null(ptr addrspace(1) dereferenceable_or_null(128) %ptr) #0 {
; HSA-LABEL: @kern_global_ptr_dereferencable_or_null(
; HSA-NEXT:    [[KERN_GLOBAL_PTR_DEREFERENCABLE_OR_NULL_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_GLOBAL_PTR_DEREFERENCABLE_OR_NULL_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !dereferenceable_or_null [[META4:![0-9]+]]
; HSA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_global_ptr_dereferencable_or_null(
; MESA-NEXT:    [[KERN_GLOBAL_PTR_DEREFERENCABLE_OR_NULL_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_GLOBAL_PTR_DEREFERENCABLE_OR_NULL_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !dereferenceable_or_null [[META4:![0-9]+]]
; MESA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; MESA-NEXT:    ret void
;
  store volatile ptr addrspace(1) %ptr, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_nonnull_global_ptr(ptr addrspace(1) nonnull %ptr) #0 {
; HSA-LABEL: @kern_nonnull_global_ptr(
; HSA-NEXT:    [[KERN_NONNULL_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_NONNULL_GLOBAL_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !nonnull [[META1]]
; HSA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_nonnull_global_ptr(
; MESA-NEXT:    [[KERN_NONNULL_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_NONNULL_GLOBAL_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !nonnull [[META1]]
; MESA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; MESA-NEXT:    ret void
;
  store volatile ptr addrspace(1) %ptr, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_align32_global_ptr(ptr addrspace(1) align 1024 %ptr) #0 {
; HSA-LABEL: @kern_align32_global_ptr(
; HSA-NEXT:    [[KERN_ALIGN32_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_ALIGN32_GLOBAL_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !align [[META5:![0-9]+]]
; HSA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_align32_global_ptr(
; MESA-NEXT:    [[KERN_ALIGN32_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_ALIGN32_GLOBAL_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !align [[META5:![0-9]+]]
; MESA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) undef, align 8
; MESA-NEXT:    ret void
;
  store volatile ptr addrspace(1) %ptr, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_noalias_global_ptr(ptr addrspace(1) noalias %ptr) #0 {
; GCN-LABEL: @kern_noalias_global_ptr(
; GCN-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; GCN-NEXT:    store volatile ptr addrspace(1) [[PTR:%.*]], ptr addrspace(1) undef, align 8
; GCN-NEXT:    ret void
;
  store volatile ptr addrspace(1) %ptr, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_noalias_global_ptr_x2(ptr addrspace(1) noalias %ptr0, ptr addrspace(1) noalias %ptr1) #0 {
; GCN-LABEL: @kern_noalias_global_ptr_x2(
; GCN-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; GCN-NEXT:    store volatile ptr addrspace(1) [[PTR0:%.*]], ptr addrspace(1) undef, align 8
; GCN-NEXT:    store volatile ptr addrspace(1) [[PTR1:%.*]], ptr addrspace(1) undef, align 8
; GCN-NEXT:    ret void
;
  store volatile ptr addrspace(1) %ptr0, ptr addrspace(1) undef
  store volatile ptr addrspace(1) %ptr1, ptr addrspace(1) undef
  ret void
}

define amdgpu_kernel void @kern_noundef_global_ptr(ptr addrspace(1) noundef %ptr) #0 {
; HSA-LABEL: @kern_noundef_global_ptr(
; HSA-NEXT:    [[KERN_NOUNDEF_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_NOUNDEF_GLOBAL_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !noundef [[META1]]
; HSA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) null, align 8
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_noundef_global_ptr(
; MESA-NEXT:    [[KERN_NOUNDEF_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[KERN_NOUNDEF_GLOBAL_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[PTR_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !noundef [[META1]]
; MESA-NEXT:    store volatile ptr addrspace(1) [[PTR_LOAD]], ptr addrspace(1) null, align 8
; MESA-NEXT:    ret void
;
  store volatile ptr addrspace(1) %ptr, ptr addrspace(1) null
  ret void
}

define amdgpu_kernel void @struct_i8_i8_arg({i8, i8} %in) #0 {
; HSA-LABEL: @struct_i8_i8_arg(
; HSA-NEXT:  entry:
; HSA-NEXT:    [[STRUCT_I8_I8_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[STRUCT_I8_I8_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN_LOAD:%.*]] = load { i8, i8 }, ptr addrspace(4) [[IN_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[ELT0:%.*]] = extractvalue { i8, i8 } [[IN_LOAD]], 0
; HSA-NEXT:    [[ELT1:%.*]] = extractvalue { i8, i8 } [[IN_LOAD]], 1
; HSA-NEXT:    store volatile i8 [[ELT0]], ptr addrspace(1) null, align 4
; HSA-NEXT:    store volatile i8 [[ELT1]], ptr addrspace(1) null, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @struct_i8_i8_arg(
; MESA-NEXT:  entry:
; MESA-NEXT:    [[STRUCT_I8_I8_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[STRUCT_I8_I8_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN_LOAD:%.*]] = load { i8, i8 }, ptr addrspace(4) [[IN_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[ELT0:%.*]] = extractvalue { i8, i8 } [[IN_LOAD]], 0
; MESA-NEXT:    [[ELT1:%.*]] = extractvalue { i8, i8 } [[IN_LOAD]], 1
; MESA-NEXT:    store volatile i8 [[ELT0]], ptr addrspace(1) null, align 4
; MESA-NEXT:    store volatile i8 [[ELT1]], ptr addrspace(1) null, align 4
; MESA-NEXT:    ret void
;
entry:
  %elt0 = extractvalue {i8, i8} %in, 0
  %elt1 = extractvalue {i8, i8} %in, 1
  store volatile i8 %elt0, ptr addrspace(1) null, align 4
  store volatile i8 %elt1, ptr addrspace(1) null, align 4
  ret void
}

define amdgpu_kernel void @struct_i8_i16_arg({i8, i16} %in) #0 {
; HSA-LABEL: @struct_i8_i16_arg(
; HSA-NEXT:  entry:
; HSA-NEXT:    [[STRUCT_I8_I16_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[STRUCT_I8_I16_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN_LOAD:%.*]] = load { i8, i16 }, ptr addrspace(4) [[IN_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[ELT0:%.*]] = extractvalue { i8, i16 } [[IN_LOAD]], 0
; HSA-NEXT:    [[ELT1:%.*]] = extractvalue { i8, i16 } [[IN_LOAD]], 1
; HSA-NEXT:    store volatile i8 [[ELT0]], ptr addrspace(1) null, align 4
; HSA-NEXT:    store volatile i16 [[ELT1]], ptr addrspace(1) null, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @struct_i8_i16_arg(
; MESA-NEXT:  entry:
; MESA-NEXT:    [[STRUCT_I8_I16_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[STRUCT_I8_I16_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN_LOAD:%.*]] = load { i8, i16 }, ptr addrspace(4) [[IN_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[ELT0:%.*]] = extractvalue { i8, i16 } [[IN_LOAD]], 0
; MESA-NEXT:    [[ELT1:%.*]] = extractvalue { i8, i16 } [[IN_LOAD]], 1
; MESA-NEXT:    store volatile i8 [[ELT0]], ptr addrspace(1) null, align 4
; MESA-NEXT:    store volatile i16 [[ELT1]], ptr addrspace(1) null, align 4
; MESA-NEXT:    ret void
;
entry:
  %elt0 = extractvalue {i8, i16} %in, 0
  %elt1 = extractvalue {i8, i16} %in, 1
  store volatile i8 %elt0, ptr addrspace(1) null, align 4
  store volatile i16 %elt1, ptr addrspace(1) null, align 4
  ret void
}

define amdgpu_kernel void @array_2xi8_arg([2 x i8] %in) #0 {
; HSA-LABEL: @array_2xi8_arg(
; HSA-NEXT:  entry:
; HSA-NEXT:    [[ARRAY_2XI8_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[ARRAY_2XI8_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN_LOAD:%.*]] = load [2 x i8], ptr addrspace(4) [[IN_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[ELT0:%.*]] = extractvalue [2 x i8] [[IN_LOAD]], 0
; HSA-NEXT:    [[ELT1:%.*]] = extractvalue [2 x i8] [[IN_LOAD]], 1
; HSA-NEXT:    store volatile i8 [[ELT0]], ptr addrspace(1) null, align 4
; HSA-NEXT:    store volatile i8 [[ELT1]], ptr addrspace(1) null, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @array_2xi8_arg(
; MESA-NEXT:  entry:
; MESA-NEXT:    [[ARRAY_2XI8_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[ARRAY_2XI8_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN_LOAD:%.*]] = load [2 x i8], ptr addrspace(4) [[IN_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[ELT0:%.*]] = extractvalue [2 x i8] [[IN_LOAD]], 0
; MESA-NEXT:    [[ELT1:%.*]] = extractvalue [2 x i8] [[IN_LOAD]], 1
; MESA-NEXT:    store volatile i8 [[ELT0]], ptr addrspace(1) null, align 4
; MESA-NEXT:    store volatile i8 [[ELT1]], ptr addrspace(1) null, align 4
; MESA-NEXT:    ret void
;
entry:
  %elt0 = extractvalue [2 x i8] %in, 0
  %elt1 = extractvalue [2 x i8] %in, 1
  store volatile i8 %elt0, ptr addrspace(1) null, align 4
  store volatile i8 %elt1, ptr addrspace(1) null, align 4
  ret void
}

define amdgpu_kernel void @array_2xi1_arg([2 x i1] %in) #0 {
; HSA-LABEL: @array_2xi1_arg(
; HSA-NEXT:  entry:
; HSA-NEXT:    [[ARRAY_2XI1_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[ARRAY_2XI1_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN_LOAD:%.*]] = load [2 x i1], ptr addrspace(4) [[IN_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[ELT0:%.*]] = extractvalue [2 x i1] [[IN_LOAD]], 0
; HSA-NEXT:    [[ELT1:%.*]] = extractvalue [2 x i1] [[IN_LOAD]], 1
; HSA-NEXT:    store volatile i1 [[ELT0]], ptr addrspace(1) null, align 4
; HSA-NEXT:    store volatile i1 [[ELT1]], ptr addrspace(1) null, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @array_2xi1_arg(
; MESA-NEXT:  entry:
; MESA-NEXT:    [[ARRAY_2XI1_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[ARRAY_2XI1_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN_LOAD:%.*]] = load [2 x i1], ptr addrspace(4) [[IN_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[ELT0:%.*]] = extractvalue [2 x i1] [[IN_LOAD]], 0
; MESA-NEXT:    [[ELT1:%.*]] = extractvalue [2 x i1] [[IN_LOAD]], 1
; MESA-NEXT:    store volatile i1 [[ELT0]], ptr addrspace(1) null, align 4
; MESA-NEXT:    store volatile i1 [[ELT1]], ptr addrspace(1) null, align 4
; MESA-NEXT:    ret void
;
entry:
  %elt0 = extractvalue [2 x i1] %in, 0
  %elt1 = extractvalue [2 x i1] %in, 1
  store volatile i1 %elt0, ptr addrspace(1) null, align 4
  store volatile i1 %elt1, ptr addrspace(1) null, align 4
  ret void
}

define amdgpu_kernel void @only_empty_struct({} %empty) #0 {
; GCN-LABEL: @only_empty_struct(
; GCN-NEXT:    [[ONLY_EMPTY_STRUCT_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(256) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; GCN-NEXT:    ret void
;
  ret void
}

define amdgpu_kernel void @empty_struct_with_other({} %empty, i32 %arg1) #0 {
; HSA-LABEL: @empty_struct_with_other(
; HSA-NEXT:    [[EMPTY_STRUCT_WITH_OTHER_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[EMPTY_STRUCT_WITH_OTHER_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store i32 [[ARG1_LOAD]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @empty_struct_with_other(
; MESA-NEXT:    [[EMPTY_STRUCT_WITH_OTHER_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[EMPTY_STRUCT_WITH_OTHER_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG1_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store i32 [[ARG1_LOAD]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  store i32 %arg1, ptr addrspace(1) undef
  ret void
}

; Should insert code after the allocas
define amdgpu_kernel void @static_alloca_kern_i32(i32 %arg0) {
; HSA-LABEL: @static_alloca_kern_i32(
; HSA-NEXT:    [[ALLOCA:%.*]] = alloca i32, align 4, addrspace(5)
; HSA-NEXT:    [[STATIC_ALLOCA_KERN_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[STATIC_ALLOCA_KERN_I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    store volatile i32 [[ARG0_LOAD]], ptr addrspace(5) [[ALLOCA]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @static_alloca_kern_i32(
; MESA-NEXT:    [[ALLOCA:%.*]] = alloca i32, align 4, addrspace(5)
; MESA-NEXT:    [[STATIC_ALLOCA_KERN_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[STATIC_ALLOCA_KERN_I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    store volatile i32 [[ARG0_LOAD]], ptr addrspace(5) [[ALLOCA]], align 4
; MESA-NEXT:    ret void
;
  %alloca = alloca i32, addrspace(5)
  store volatile i32 %arg0, ptr addrspace(5) %alloca
  ret void
}

; Make sure we don't break the IR if an alloca depends on the
; kernargs.
define amdgpu_kernel void @dyn_alloca_kernarg_i32(i32 %n) {
; HSA-LABEL: @dyn_alloca_kernarg_i32(
; HSA-NEXT:    [[ALLOCA0:%.*]] = alloca i32, align 4, addrspace(5)
; HSA-NEXT:    [[DYN_ALLOCA_KERNARG_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[N_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[DYN_ALLOCA_KERNARG_I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[N_LOAD:%.*]] = load i32, ptr addrspace(4) [[N_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[ALLOCA1:%.*]] = alloca i32, i32 [[N_LOAD]], align 4, addrspace(5)
; HSA-NEXT:    store volatile i32 0, ptr addrspace(5) [[ALLOCA0]], align 4
; HSA-NEXT:    store volatile i32 1, ptr addrspace(5) [[ALLOCA1]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @dyn_alloca_kernarg_i32(
; MESA-NEXT:    [[ALLOCA0:%.*]] = alloca i32, align 4, addrspace(5)
; MESA-NEXT:    [[DYN_ALLOCA_KERNARG_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[N_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[DYN_ALLOCA_KERNARG_I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[N_LOAD:%.*]] = load i32, ptr addrspace(4) [[N_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[ALLOCA1:%.*]] = alloca i32, i32 [[N_LOAD]], align 4, addrspace(5)
; MESA-NEXT:    store volatile i32 0, ptr addrspace(5) [[ALLOCA0]], align 4
; MESA-NEXT:    store volatile i32 1, ptr addrspace(5) [[ALLOCA1]], align 4
; MESA-NEXT:    ret void
;
  %alloca0 = alloca i32, addrspace(5)
  %alloca1 = alloca i32, i32 %n, addrspace(5)
  store volatile i32 0, ptr addrspace(5) %alloca0
  store volatile i32 1, ptr addrspace(5) %alloca1
  ret void
}

; Byref pointers should only be treated as offsets from kernarg
define amdgpu_kernel void @byref_constant_i8_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i8) %in.byref) {
; HSA-LABEL: @byref_constant_i8_arg(
; HSA-NEXT:    [[BYREF_CONSTANT_I8_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I8_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I8_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[IN:%.*]] = load i8, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 1
; HSA-NEXT:    [[EXT:%.*]] = zext i8 [[IN]] to i32
; HSA-NEXT:    store i32 [[EXT]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_constant_i8_arg(
; MESA-NEXT:    [[BYREF_CONSTANT_I8_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(268) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I8_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I8_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[IN:%.*]] = load i8, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 1
; MESA-NEXT:    [[EXT:%.*]] = zext i8 [[IN]] to i32
; MESA-NEXT:    store i32 [[EXT]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i8, ptr addrspace(4) %in.byref
  %ext = zext i8 %in to i32
  store i32 %ext, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_constant_i16_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i16) %in.byref) {
; HSA-LABEL: @byref_constant_i16_arg(
; HSA-NEXT:    [[BYREF_CONSTANT_I16_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I16_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I16_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[IN:%.*]] = load i16, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 2
; HSA-NEXT:    [[EXT:%.*]] = zext i16 [[IN]] to i32
; HSA-NEXT:    store i32 [[EXT]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_constant_i16_arg(
; MESA-NEXT:    [[BYREF_CONSTANT_I16_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(268) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I16_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I16_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[IN:%.*]] = load i16, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 2
; MESA-NEXT:    [[EXT:%.*]] = zext i16 [[IN]] to i32
; MESA-NEXT:    store i32 [[EXT]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i16, ptr addrspace(4) %in.byref
  %ext = zext i16 %in to i32
  store i32 %ext, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_constant_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i32) %in.byref, i32 %after.offset) {
; HSA-LABEL: @byref_constant_i32_arg(
; HSA-NEXT:    [[BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 12
; HSA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; HSA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; HSA-NEXT:    store volatile i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_constant_i32_arg(
; MESA-NEXT:    [[BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 48
; MESA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; MESA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; MESA-NEXT:    store volatile i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i32, ptr addrspace(4) %in.byref
  store volatile i32 %in, ptr addrspace(1) %out, align 4
  store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_constant_v4i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(<4 x i32>) %in.byref, i32 %after.offset) {
; HSA-LABEL: @byref_constant_v4i32_arg(
; HSA-NEXT:    [[BYREF_CONSTANT_V4I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(296) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_V4I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_V4I32_ARG_KERNARG_SEGMENT]], i64 16
; HSA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_V4I32_ARG_KERNARG_SEGMENT]], i64 32
; HSA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN:%.*]] = load <4 x i32>, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 16
; HSA-NEXT:    store volatile <4 x i32> [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_constant_v4i32_arg(
; MESA-NEXT:    [[BYREF_CONSTANT_V4I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(292) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_V4I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_V4I32_ARG_KERNARG_SEGMENT]], i64 52
; MESA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_V4I32_ARG_KERNARG_SEGMENT]], i64 68
; MESA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN:%.*]] = load <4 x i32>, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 16
; MESA-NEXT:    store volatile <4 x i32> [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load <4 x i32>, ptr addrspace(4) %in.byref
  store volatile <4 x i32> %in, ptr addrspace(1) %out, align 4
  store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_align_constant_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i32) align(256) %in.byref, i32 %after.offset) {
; HSA-LABEL: @byref_align_constant_i32_arg(
; HSA-NEXT:    [[BYREF_ALIGN_CONSTANT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 256 dereferenceable(520) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_ALIGN_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_ALIGN_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 256
; HSA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_ALIGN_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 260
; HSA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; HSA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; HSA-NEXT:    store volatile i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_align_constant_i32_arg(
; MESA-NEXT:    [[BYREF_ALIGN_CONSTANT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 256 dereferenceable(520) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_ALIGN_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_ALIGN_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 292
; MESA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_ALIGN_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 296
; MESA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 8, !invariant.load [[META1]]
; MESA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; MESA-NEXT:    store volatile i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i32, ptr addrspace(4) %in.byref
  store volatile i32 %in, ptr addrspace(1) %out, align 4
  store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_natural_align_constant_v16i32_arg(ptr addrspace(1) nocapture %out, i8, ptr addrspace(4) byref(<16 x i32>) %in.byref, i32 %after.offset) {
; HSA-LABEL: @byref_natural_align_constant_v16i32_arg(
; HSA-NEXT:    [[BYREF_NATURAL_ALIGN_CONSTANT_V16I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 64 dereferenceable(392) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_NATURAL_ALIGN_CONSTANT_V16I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_NATURAL_ALIGN_CONSTANT_V16I32_ARG_KERNARG_SEGMENT]], i64 64
; HSA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_NATURAL_ALIGN_CONSTANT_V16I32_ARG_KERNARG_SEGMENT]], i64 128
; HSA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN:%.*]] = load <16 x i32>, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 64
; HSA-NEXT:    store volatile <16 x i32> [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_natural_align_constant_v16i32_arg(
; MESA-NEXT:    [[BYREF_NATURAL_ALIGN_CONSTANT_V16I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 64 dereferenceable(388) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_NATURAL_ALIGN_CONSTANT_V16I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_NATURAL_ALIGN_CONSTANT_V16I32_ARG_KERNARG_SEGMENT]], i64 100
; MESA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_NATURAL_ALIGN_CONSTANT_V16I32_ARG_KERNARG_SEGMENT]], i64 164
; MESA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN:%.*]] = load <16 x i32>, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 64
; MESA-NEXT:    store volatile <16 x i32> [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load <16 x i32>, ptr addrspace(4) %in.byref
  store volatile <16 x i32> %in, ptr addrspace(1) %out, align 4
  store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
  ret void
}

; Also accept byref kernel arguments with other global address spaces.
define amdgpu_kernel void @byref_global_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(1) byref(i32) %in.byref) {
; HSA-LABEL: @byref_global_i32_arg(
; HSA-NEXT:    [[BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr addrspace(1)
; HSA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(1) [[TMP1]], align 4
; HSA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_global_i32_arg(
; MESA-NEXT:    [[BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(268) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr addrspace(1)
; MESA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(1) [[TMP1]], align 4
; MESA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i32, ptr addrspace(1) %in.byref
  store i32 %in, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_flat_i32_arg(ptr addrspace(1) nocapture %out, ptr byref(i32) %in.byref) {
; HSA-LABEL: @byref_flat_i32_arg(
; HSA-NEXT:    [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr
; HSA-NEXT:    [[IN:%.*]] = load i32, ptr [[TMP1]], align 4
; HSA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_flat_i32_arg(
; MESA-NEXT:    [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(268) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr
; MESA-NEXT:    [[IN:%.*]] = load i32, ptr [[TMP1]], align 4
; MESA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i32, ptr %in.byref
  store i32 %in, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_constant_32bit_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(6) byref(i32) %in.byref) {
; HSA-LABEL: @byref_constant_32bit_i32_arg(
; HSA-NEXT:    [[BYREF_CONSTANT_32BIT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_32BIT_I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_32BIT_I32_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr addrspace(6)
; HSA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(6) [[TMP1]], align 4
; HSA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_constant_32bit_i32_arg(
; MESA-NEXT:    [[BYREF_CONSTANT_32BIT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(268) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_32BIT_I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_32BIT_I32_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr addrspace(6)
; MESA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(6) [[TMP1]], align 4
; MESA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i32, ptr addrspace(6) %in.byref
  store i32 %in, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_unknown_as_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(999) byref(i32) %in.byref) {
; HSA-LABEL: @byref_unknown_as_i32_arg(
; HSA-NEXT:    [[BYREF_UNKNOWN_AS_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_UNKNOWN_AS_I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_UNKNOWN_AS_I32_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr addrspace(999)
; HSA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(999) [[TMP1]], align 4
; HSA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_unknown_as_i32_arg(
; MESA-NEXT:    [[BYREF_UNKNOWN_AS_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(268) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_UNKNOWN_AS_I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_UNKNOWN_AS_I32_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr addrspace(999)
; MESA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(999) [[TMP1]], align 4
; MESA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i32, ptr addrspace(999) %in.byref
  store i32 %in, ptr addrspace(1) %out, align 4
  ret void
}

; Invalid, but should not crash.
define amdgpu_kernel void @byref_local_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(3) byref(i32) %in.byref) {
; HSA-LABEL: @byref_local_i32_arg(
; HSA-NEXT:    [[BYREF_LOCAL_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_LOCAL_I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_LOCAL_I32_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr addrspace(3)
; HSA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(3) [[TMP1]], align 4
; HSA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_local_i32_arg(
; MESA-NEXT:    [[BYREF_LOCAL_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(268) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_LOCAL_I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_LOCAL_I32_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to ptr addrspace(3)
; MESA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(3) [[TMP1]], align 4
; MESA-NEXT:    store i32 [[IN]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in = load i32, ptr addrspace(3) %in.byref
  store i32 %in, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @multi_byref_constant_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i32) %in0.byref, ptr addrspace(4) byref(i32) %in1.byref, i32 %after.offset) {
; HSA-LABEL: @multi_byref_constant_i32_arg(
; HSA-NEXT:    [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(280) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN0_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 8
; HSA-NEXT:    [[IN1_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 12
; HSA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 16
; HSA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 16, !invariant.load [[META1]]
; HSA-NEXT:    [[IN0:%.*]] = load i32, ptr addrspace(4) [[IN0_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; HSA-NEXT:    [[IN1:%.*]] = load i32, ptr addrspace(4) [[IN1_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; HSA-NEXT:    store volatile i32 [[IN0]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    store volatile i32 [[IN1]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @multi_byref_constant_i32_arg(
; MESA-NEXT:    [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(276) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[OUT_LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(4) [[OUT_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN0_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 44
; MESA-NEXT:    [[IN1_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 48
; MESA-NEXT:    [[AFTER_OFFSET_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[MULTI_BYREF_CONSTANT_I32_ARG_KERNARG_SEGMENT]], i64 52
; MESA-NEXT:    [[AFTER_OFFSET_LOAD:%.*]] = load i32, ptr addrspace(4) [[AFTER_OFFSET_KERNARG_OFFSET]], align 4, !invariant.load [[META1]]
; MESA-NEXT:    [[IN0:%.*]] = load i32, ptr addrspace(4) [[IN0_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; MESA-NEXT:    [[IN1:%.*]] = load i32, ptr addrspace(4) [[IN1_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; MESA-NEXT:    store volatile i32 [[IN0]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    store volatile i32 [[IN1]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    store volatile i32 [[AFTER_OFFSET_LOAD]], ptr addrspace(1) [[OUT_LOAD]], align 4
; MESA-NEXT:    ret void
;
  %in0 = load i32, ptr addrspace(4) %in0.byref
  %in1 = load i32, ptr addrspace(4) %in1.byref
  store volatile i32 %in0, ptr addrspace(1) %out, align 4
  store volatile i32 %in1, ptr addrspace(1) %out, align 4
  store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @byref_constant_i32_arg_offset0(ptr addrspace(4) byref(i32) %in.byref) {
; HSA-LABEL: @byref_constant_i32_arg_offset0(
; HSA-NEXT:    [[BYREF_CONSTANT_I32_ARG_OFFSET0_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I32_ARG_OFFSET0_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; HSA-NEXT:    store i32 [[IN]], ptr addrspace(1) undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @byref_constant_i32_arg_offset0(
; MESA-NEXT:    [[BYREF_CONSTANT_I32_ARG_OFFSET0_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[BYREF_CONSTANT_I32_ARG_OFFSET0_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN:%.*]] = load i32, ptr addrspace(4) [[IN_BYREF_BYVAL_KERNARG_OFFSET]], align 4
; MESA-NEXT:    store i32 [[IN]], ptr addrspace(1) undef, align 4
; MESA-NEXT:    ret void
;
  %in = load i32, ptr addrspace(4) %in.byref
  store i32 %in, ptr addrspace(1) undef, align 4
  ret void
}

define amdgpu_kernel void @noundef_f32(float noundef %arg0) {
; HSA-LABEL: @noundef_f32(
; HSA-NEXT:    [[NOUNDEF_F32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_F32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load float, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !noundef [[META1]]
; HSA-NEXT:    call void (...) @llvm.fake.use(float [[ARG0_LOAD]])
; HSA-NEXT:    ret void
;
; MESA-LABEL: @noundef_f32(
; MESA-NEXT:    [[NOUNDEF_F32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_F32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load float, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !noundef [[META1]]
; MESA-NEXT:    call void (...) @llvm.fake.use(float [[ARG0_LOAD]])
; MESA-NEXT:    ret void
;
  call void (...) @llvm.fake.use(float %arg0)
  ret void
}

define amdgpu_kernel void @noundef_f16(half noundef %arg0) {
; HSA-LABEL: @noundef_f16(
; HSA-NEXT:    [[NOUNDEF_F16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_F16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 16, !invariant.load [[META1]], !noundef [[META1]]
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = bitcast i16 [[TMP2]] to half
; HSA-NEXT:    call void (...) @llvm.fake.use(half [[ARG0_LOAD]])
; HSA-NEXT:    ret void
;
; MESA-LABEL: @noundef_f16(
; MESA-NEXT:    [[NOUNDEF_F16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(260) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_F16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(4) [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]], align 4, !invariant.load [[META1]], !noundef [[META1]]
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = bitcast i16 [[TMP2]] to half
; MESA-NEXT:    call void (...) @llvm.fake.use(half [[ARG0_LOAD]])
; MESA-NEXT:    ret void
;
  call void (...) @llvm.fake.use(half %arg0)
  ret void
}

define amdgpu_kernel void @noundef_v2i32(<2 x i32> noundef %arg0) {
; HSA-LABEL: @noundef_v2i32(
; HSA-NEXT:    [[NOUNDEF_V2I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_V2I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load <2 x i32>, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !noundef [[META1]]
; HSA-NEXT:    call void (...) @llvm.fake.use(<2 x i32> [[ARG0_LOAD]])
; HSA-NEXT:    ret void
;
; MESA-LABEL: @noundef_v2i32(
; MESA-NEXT:    [[NOUNDEF_V2I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_V2I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load <2 x i32>, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !noundef [[META1]]
; MESA-NEXT:    call void (...) @llvm.fake.use(<2 x i32> [[ARG0_LOAD]])
; MESA-NEXT:    ret void
;
  call void (...) @llvm.fake.use(<2 x i32> %arg0)
  ret void
}

define amdgpu_kernel void @noundef_p0(ptr noundef %arg0) {
; HSA-LABEL: @noundef_p0(
; HSA-NEXT:    [[NOUNDEF_P0_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_P0_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load ptr, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !noundef [[META1]]
; HSA-NEXT:    call void (...) @llvm.fake.use(ptr [[ARG0_LOAD]])
; HSA-NEXT:    ret void
;
; MESA-LABEL: @noundef_p0(
; MESA-NEXT:    [[NOUNDEF_P0_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_P0_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load ptr, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !noundef [[META1]]
; MESA-NEXT:    call void (...) @llvm.fake.use(ptr [[ARG0_LOAD]])
; MESA-NEXT:    ret void
;
  call void (...) @llvm.fake.use(ptr %arg0)
  ret void
}

define amdgpu_kernel void @noundef_v2p0(<2 x ptr> noundef %arg0) {
; HSA-LABEL: @noundef_v2p0(
; HSA-NEXT:    [[NOUNDEF_V2P0_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_V2P0_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load <2 x ptr>, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 16, !invariant.load [[META1]], !noundef [[META1]]
; HSA-NEXT:    call void (...) @llvm.fake.use(<2 x ptr> [[ARG0_LOAD]])
; HSA-NEXT:    ret void
;
; MESA-LABEL: @noundef_v2p0(
; MESA-NEXT:    [[NOUNDEF_V2P0_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[NOUNDEF_V2P0_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load <2 x ptr>, ptr addrspace(4) [[ARG0_KERNARG_OFFSET]], align 4, !invariant.load [[META1]], !noundef [[META1]]
; MESA-NEXT:    call void (...) @llvm.fake.use(<2 x ptr> [[ARG0_LOAD]])
; MESA-NEXT:    ret void
;
  call void (...) @llvm.fake.use(<2 x ptr> %arg0)
  ret void
}

attributes #0 = { nounwind "target-cpu"="kaveri" }
attributes #1 = { nounwind "target-cpu"="kaveri" "amdgpu-implicitarg-num-bytes"="40" }
attributes #2 = { nounwind "target-cpu"="tahiti" }


!llvm.module.flags = !{!0}
!0 = !{i32 1, !"amdhsa_code_object_version", i32 500}
;.
; HSA: attributes #[[ATTR0:[0-9]+]] = { nounwind }
; HSA: attributes #[[ATTR1:[0-9]+]] = { nounwind "target-cpu"="kaveri" }
; HSA: attributes #[[ATTR2:[0-9]+]] = { nounwind "amdgpu-implicitarg-num-bytes"="40" "target-cpu"="kaveri" }
; HSA: attributes #[[ATTR3:[0-9]+]] = { nounwind "target-cpu"="tahiti" }
; HSA: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
;.
; MESA: attributes #[[ATTR0:[0-9]+]] = { nounwind }
; MESA: attributes #[[ATTR1:[0-9]+]] = { nounwind "target-cpu"="kaveri" }
; MESA: attributes #[[ATTR2:[0-9]+]] = { nounwind "amdgpu-implicitarg-num-bytes"="40" "target-cpu"="kaveri" }
; MESA: attributes #[[ATTR3:[0-9]+]] = { nounwind "target-cpu"="tahiti" }
; MESA: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
;.
; HSA: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 500}
; HSA: [[META1]] = !{}
; HSA: [[RNG2]] = !{i32 0, i32 8}
; HSA: [[META3]] = !{i64 42}
; HSA: [[META4]] = !{i64 128}
; HSA: [[META5]] = !{i64 1024}
;.
; MESA: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 500}
; MESA: [[META1]] = !{}
; MESA: [[RNG2]] = !{i32 0, i32 8}
; MESA: [[META3]] = !{i64 42}
; MESA: [[META4]] = !{i64 128}
; MESA: [[META5]] = !{i64 1024}
;.