; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; FIXME: Manually added checks for metadata nodes at bottom
; RUN: opt -mtriple=amdgcn-amd-amdhsa -S -o - -amdgpu-lower-kernel-arguments %s | FileCheck -check-prefix=HSA %s
; RUN: opt -mtriple=amdgcn-- -S -o - -amdgpu-lower-kernel-arguments %s | FileCheck -check-prefix=MESA %s

define amdgpu_kernel void @kern_noargs() {
; HSA-LABEL: @kern_noargs(
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_noargs(
; MESA-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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    store i8 [[TMP2]], i8 addrspace(1)* undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i8(
; MESA-NEXT:    [[KERN_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    store i8 [[TMP2]], i8 addrspace(1)* undef, align 1
; MESA-NEXT:    ret void
;
  store i8 %arg, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    store i16 [[TMP2]], i16 addrspace(1)* undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i16(
; MESA-NEXT:    [[KERN_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    store i16 [[TMP2]], i16 addrspace(1)* undef, align 1
; MESA-NEXT:    ret void
;
  store i16 %arg, i16 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_F16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    [[ARG_LOAD:%.*]] = bitcast i16 [[TMP2]] to half
; HSA-NEXT:    store half [[ARG_LOAD]], half addrspace(1)* undef, align 1
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_f16(
; MESA-NEXT:    [[KERN_F16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_F16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    [[ARG_LOAD:%.*]] = bitcast i16 [[TMP2]] to half
; MESA-NEXT:    store half [[ARG_LOAD]], half addrspace(1)* undef, align 1
; MESA-NEXT:    ret void
;
  store half %arg, half 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_ZEROEXT_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    store i8 [[TMP2]], i8 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_ZEROEXT_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    store i8 [[TMP2]], i8 addrspace(1)* undef, align 1
; MESA-NEXT:    ret void
;
  store i8 %arg, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_ZEROEXT_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    store i16 [[TMP2]], i16 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_ZEROEXT_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    store i16 [[TMP2]], i16 addrspace(1)* undef, align 1
; MESA-NEXT:    ret void
;
  store i16 %arg, i16 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_SIGNEXT_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    store i8 [[TMP2]], i8 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_SIGNEXT_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    store i8 [[TMP2]], i8 addrspace(1)* undef, align 1
; MESA-NEXT:    ret void
;
  store i8 %arg, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_SIGNEXT_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; HSA-NEXT:    store i16 [[TMP2]], i16 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_SIGNEXT_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
; MESA-NEXT:    store i16 [[TMP2]], i16 addrspace(1)* undef, align 1
; MESA-NEXT:    ret void
;
  store i16 %arg, i16 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef, align 1
; HSA-NEXT:    store volatile i8 [[TMP5]], i8 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef, align 1
; MESA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef, align 1
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, i8 addrspace(1)* undef, align 1
  store volatile i8 %arg1, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V3I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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]], <3 x i8> addrspace(1)* undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v3i8(
; MESA-NEXT:    [[KERN_V3I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V3I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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]], <3 x i8> addrspace(1)* undef, align 4
; MESA-NEXT:    ret void
;
  store <3 x i8> %arg, <3 x i8> 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I24_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i24
; HSA-NEXT:    store i24 [[TMP2]], i24 addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i24(
; MESA-NEXT:    [[KERN_I24_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I24_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i24
; MESA-NEXT:    store i24 [[TMP2]], i24 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store i24 %arg0, i24 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to i32 addrspace(4)*
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store i32 [[ARG0_LOAD]], i32 addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_i32(
; MESA-NEXT:    [[KERN_I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to i32 addrspace(4)*
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store i32 [[ARG0_LOAD]], i32 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store i32 %arg0, i32 addrspace(1)* undef
  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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_F32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to float addrspace(4)*
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load float, float addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store float [[ARG0_LOAD]], float addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_f32(
; MESA-NEXT:    [[KERN_F32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_F32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to float addrspace(4)*
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load float, float addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store float [[ARG0_LOAD]], float addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store float %arg0, float 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(16) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V3I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to <4 x i32> addrspace(4)*
; HSA-NEXT:    [[TMP:%.*]] = load <4 x i32>, <4 x i32> addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = shufflevector <4 x i32> [[TMP]], <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
; HSA-NEXT:    store <3 x i32> [[ARG0_LOAD]], <3 x i32> addrspace(1)* undef, align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v3i32(
; MESA-NEXT:    [[KERN_V3I32_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(52) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V3I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to <4 x i32> addrspace(4)*
; MESA-NEXT:    [[TMP:%.*]] = load <4 x i32>, <4 x i32> addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = shufflevector <4 x i32> [[TMP]], <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
; MESA-NEXT:    store <3 x i32> [[ARG0_LOAD]], <3 x i32> addrspace(1)* undef, align 4
; MESA-NEXT:    ret void
;
  store <3 x i32> %arg0, <3 x i32> 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(32) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V8I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET]] to <8 x i32> addrspace(4)*
; HSA-NEXT:    [[ARG_LOAD:%.*]] = load <8 x i32>, <8 x i32> addrspace(4)* [[ARG_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store <8 x i32> [[ARG_LOAD]], <8 x i32> addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v8i32(
; MESA-NEXT:    [[KERN_V8I32_KERNARG_SEGMENT:%.*]] = call nonnull align 32 dereferenceable(68) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V8I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET]] to <8 x i32> addrspace(4)*
; MESA-NEXT:    [[ARG_LOAD:%.*]] = load <8 x i32>, <8 x i32> addrspace(4)* [[ARG_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store <8 x i32> [[ARG_LOAD]], <8 x i32> addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store <8 x i32> %arg, <8 x i32> 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(64) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V8I64_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET]] to <8 x i64> addrspace(4)*
; HSA-NEXT:    [[ARG_LOAD:%.*]] = load <8 x i64>, <8 x i64> addrspace(4)* [[ARG_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store <8 x i64> [[ARG_LOAD]], <8 x i64> addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v8i64(
; MESA-NEXT:    [[KERN_V8I64_KERNARG_SEGMENT:%.*]] = call nonnull align 64 dereferenceable(100) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V8I64_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET]] to <8 x i64> addrspace(4)*
; MESA-NEXT:    [[ARG_LOAD:%.*]] = load <8 x i64>, <8 x i64> addrspace(4)* [[ARG_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store <8 x i64> [[ARG_LOAD]], <8 x i64> addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store <8 x i64> %arg, <8 x i64> 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(128) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V16I64_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET]] to <16 x i64> addrspace(4)*
; HSA-NEXT:    [[ARG_LOAD:%.*]] = load <16 x i64>, <16 x i64> addrspace(4)* [[ARG_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store <16 x i64> [[ARG_LOAD]], <16 x i64> addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_v16i64(
; MESA-NEXT:    [[KERN_V16I64_KERNARG_SEGMENT:%.*]] = call nonnull align 128 dereferenceable(164) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_V16I64_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG_KERNARG_OFFSET]] to <16 x i64> addrspace(4)*
; MESA-NEXT:    [[ARG_LOAD:%.*]] = load <16 x i64>, <16 x i64> addrspace(4)* [[ARG_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store <16 x i64> [[ARG_LOAD]], <16 x i64> addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store <16 x i64> %arg, <16 x i64> 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(32) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I32_V3I32_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to i32 addrspace(4)*
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I32_V3I32_KERNARG_SEGMENT]], i64 16
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET]] to <4 x i32> addrspace(4)*
; HSA-NEXT:    [[TMP:%.*]] = load <4 x i32>, <4 x i32> addrspace(4)* [[ARG1_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = shufflevector <4 x i32> [[TMP]], <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
; HSA-NEXT:    store i32 [[ARG0_LOAD]], i32 addrspace(1)* undef
; HSA-NEXT:    store <3 x i32> [[ARG1_LOAD]], <3 x i32> 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(68) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I32_V3I32_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to i32 addrspace(4)*
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_I32_V3I32_KERNARG_SEGMENT]], i64 52
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET]] to <4 x i32> addrspace(4)*
; MESA-NEXT:    [[TMP:%.*]] = load <4 x i32>, <4 x i32> addrspace(4)* [[ARG1_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = shufflevector <4 x i32> [[TMP]], <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
; MESA-NEXT:    store i32 [[ARG0_LOAD]], i32 addrspace(1)* undef
; MESA-NEXT:    store <3 x i32> [[ARG1_LOAD]], <3 x i32> addrspace(1)* undef, align 4
; MESA-NEXT:    ret void
;
  store i32 %arg0, i32 addrspace(1)* undef
  store <3 x i32> %arg1, <3 x i32> 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(12) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_STRUCT_A_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to [[STRUCT_A:%.*]] addrspace(4)*
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load [[STRUCT_A]], [[STRUCT_A]] addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store [[STRUCT_A]] %arg0.load, [[STRUCT_A]] addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_struct_a(
; MESA-NEXT:    [[KERN_STRUCT_A_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(48) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_STRUCT_A_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to [[STRUCT_A:%.*]] addrspace(4)*
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load [[STRUCT_A]], [[STRUCT_A]] addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store [[STRUCT_A]] %arg0.load, [[STRUCT_A]] addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store %struct.a %arg0, %struct.a 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(32) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_STRUCT_B_PACKED_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to [[STRUCT_B_PACKED:%.*]] addrspace(4)*
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load [[STRUCT_B_PACKED]], [[STRUCT_B_PACKED]] addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store [[STRUCT_B_PACKED]] %arg0.load, [[STRUCT_B_PACKED]] addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_struct_b_packed(
; MESA-NEXT:    [[KERN_STRUCT_B_PACKED_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(68) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_STRUCT_B_PACKED_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to [[STRUCT_B_PACKED:%.*]] addrspace(4)*
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load [[STRUCT_B_PACKED]], [[STRUCT_B_PACKED]] addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store [[STRUCT_B_PACKED]] %arg0.load, [[STRUCT_B_PACKED]] addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store %struct.b.packed %arg0, %struct.b.packed 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) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_IMPLICIT_ARG_NUM_BYTES_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to i32 addrspace(4)*
; HSA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store i32 [[ARG0_LOAD]], i32 addrspace(1)* undef
; 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) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_IMPLICIT_ARG_NUM_BYTES_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET]] to i32 addrspace(4)*
; MESA-NEXT:    [[ARG0_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store i32 [[ARG0_LOAD]], i32 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store i32 %arg0, i32 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) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERNEL_IMPLICITARG_NO_STRUCT_ALIGN_KERNARG_SEGMENT]], i64 64
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET]] to i32 addrspace(4)*
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store i32 [[ARG1_LOAD]], i32 addrspace(1)* undef
; 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) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERNEL_IMPLICITARG_NO_STRUCT_ALIGN_KERNARG_SEGMENT]], i64 100
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET]] to i32 addrspace(4)*
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store i32 [[ARG1_LOAD]], i32 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store i32 %arg1, i32 addrspace(1)* undef
  ret void
}

define amdgpu_kernel void @kern_lds_ptr(i32 addrspace(3)* %lds) #0 {
; HSA-LABEL: @kern_lds_ptr(
; HSA-NEXT:    [[KERN_LDS_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[LDS_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_LDS_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[LDS_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[LDS_KERNARG_OFFSET]] to i32 addrspace(3)* addrspace(4)*
; HSA-NEXT:    [[LDS_LOAD:%.*]] = load i32 addrspace(3)*, i32 addrspace(3)* addrspace(4)* [[LDS_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store i32 0, i32 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(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[LDS_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_LDS_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[LDS_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[LDS_KERNARG_OFFSET]] to i32 addrspace(3)* addrspace(4)*
; MESA-NEXT:    [[LDS_LOAD:%.*]] = load i32 addrspace(3)*, i32 addrspace(3)* addrspace(4)* [[LDS_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store i32 0, i32 addrspace(3)* [[LDS_LOAD]], align 4
; MESA-NEXT:    ret void
;
  store i32 0, i32 addrspace(3)* %lds, align 4
  ret void
}

define amdgpu_kernel void @kern_lds_ptr_si(i32 addrspace(3)* %lds) #2 {
; HSA-LABEL: @kern_lds_ptr_si(
; HSA-NEXT:    [[KERN_LDS_PTR_SI_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[LDS_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_LDS_PTR_SI_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[LDS_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[LDS_KERNARG_OFFSET]] to i32 addrspace(3)* addrspace(4)*
; HSA-NEXT:    [[LDS_LOAD:%.*]] = load i32 addrspace(3)*, i32 addrspace(3)* addrspace(4)* [[LDS_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store i32 0, i32 addrspace(3)* [[LDS_LOAD]], align 4
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_lds_ptr_si(
; MESA-NEXT:    [[KERN_LDS_PTR_SI_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    store i32 0, i32 addrspace(3)* [[LDS:%.*]], align 4
; MESA-NEXT:    ret void
;
  store i32 0, i32 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_i8(
; MESA-NEXT:    [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, i8 addrspace(1)* undef
  store volatile i8 %arg1, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; HSA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP8]], i8 addrspace(1)* undef
; 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; MESA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP8]], i8 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, i8 addrspace(1)* undef
  store volatile i8 %arg1, i8 addrspace(1)* undef
  store volatile i8 %arg2, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP9:%.*]] = load i32, i32 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; HSA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP8]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP11]], i8 addrspace(1)* undef
; 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP9:%.*]] = load i32, i32 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; MESA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP8]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP11]], i8 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, i8 addrspace(1)* undef
  store volatile i8 %arg1, i8 addrspace(1)* undef
  store volatile i8 %arg2, i8 addrspace(1)* undef
  store volatile i8 %arg3, i8 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(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile <3 x i8> [[ARG1_LOAD]], <3 x i8> addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_v3i8(
; MESA-NEXT:    [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_V3I8_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 8, !invariant.load !0
; 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]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile <3 x i8> [[ARG1_LOAD]], <3 x i8> addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, i8 addrspace(1)* undef
  store volatile <3 x i8> %arg1, <3 x i8> 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; HSA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i16 [[TMP5]], i16 addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i8_i16(
; MESA-NEXT:    [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; MESA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i16 [[TMP5]], i16 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, i8 addrspace(1)* undef
  store volatile i16 %arg1, i16 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i1
; HSA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; HSA-NEXT:    store volatile i1 [[TMP5]], i1 addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i1_i1(
; MESA-NEXT:    [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 8
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i1
; MESA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; MESA-NEXT:    store volatile i1 [[TMP5]], i1 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, i1 addrspace(1)* undef
  store volatile i1 %arg1, i1 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; HSA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i1
; HSA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; HSA-NEXT:    store volatile i1 [[TMP5]], i1 addrspace(1)* undef
; HSA-NEXT:    store volatile i1 [[TMP8]], i1 addrspace(1)* undef
; 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP7:%.*]] = lshr i32 [[TMP6]], 16
; MESA-NEXT:    [[TMP8:%.*]] = trunc i32 [[TMP7]] to i1
; MESA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; MESA-NEXT:    store volatile i1 [[TMP5]], i1 addrspace(1)* undef
; MESA-NEXT:    store volatile i1 [[TMP8]], i1 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, i1 addrspace(1)* undef
  store volatile i1 %arg1, i1 addrspace(1)* undef
  store volatile i1 %arg2, i1 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP9:%.*]] = load i32, i32 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; HSA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i1
; HSA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; HSA-NEXT:    store volatile i1 [[TMP5]], i1 addrspace(1)* undef
; HSA-NEXT:    store volatile i1 [[TMP8]], i1 addrspace(1)* undef
; HSA-NEXT:    store volatile i1 [[TMP11]], i1 addrspace(1)* undef
; 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I1_I1_I1_I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP9:%.*]] = load i32, i32 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP10:%.*]] = lshr i32 [[TMP9]], 24
; MESA-NEXT:    [[TMP11:%.*]] = trunc i32 [[TMP10]] to i1
; MESA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; MESA-NEXT:    store volatile i1 [[TMP5]], i1 addrspace(1)* undef
; MESA-NEXT:    store volatile i1 [[TMP8]], i1 addrspace(1)* undef
; MESA-NEXT:    store volatile i1 [[TMP11]], i1 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, i1 addrspace(1)* undef
  store volatile i1 %arg1, i1 addrspace(1)* undef
  store volatile i1 %arg2, i1 addrspace(1)* undef
  store volatile i1 %arg3, i1 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(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; HSA-NEXT:    [[TMP4:%.*]] = trunc i32 [[TMP3]] to i3
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = bitcast i3 [[TMP4]] to <3 x i1>
; HSA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; HSA-NEXT:    store volatile <3 x i1> [[ARG1_LOAD]], <3 x i1> addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i1_v3i1(
; MESA-NEXT:    [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_V3I1_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 8, !invariant.load !0
; MESA-NEXT:    [[TMP4:%.*]] = trunc i32 [[TMP3]] to i3
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = bitcast i3 [[TMP4]] to <3 x i1>
; MESA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; MESA-NEXT:    store volatile <3 x i1> [[ARG1_LOAD]], <3 x i1> addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, i1 addrspace(1)* undef
  store volatile <3 x i1> %arg1, <3 x i1> 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; HSA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; HSA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; HSA-NEXT:    store volatile i16 [[TMP5]], i16 addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_i1_i16(
; MESA-NEXT:    [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i1
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I1_I16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP3]], 16
; MESA-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
; MESA-NEXT:    store volatile i1 [[TMP2]], i1 addrspace(1)* undef
; MESA-NEXT:    store volatile i16 [[TMP5]], i16 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i1 %arg0, i1 addrspace(1)* undef
  store volatile i16 %arg1, i16 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(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP9:%.*]] = load i32, i32 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[ARG5_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG5_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP12:%.*]] = load i32, i32 addrspace(4)* [[ARG5_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[ARG6_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG6_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP15:%.*]] = load i32, i32 addrspace(4)* [[ARG6_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 4
; HSA-NEXT:    [[ARG7_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG7_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP18:%.*]] = load i32, i32 addrspace(4)* [[ARG7_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; HSA-NEXT:    [[TMP19:%.*]] = lshr i32 [[TMP18]], 24
; HSA-NEXT:    [[TMP20:%.*]] = trunc i32 [[TMP19]] to i8
; HSA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP8]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP11]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP14]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP17]], i8 addrspace(1)* undef
; HSA-NEXT:    store volatile i8 [[TMP20]], i8 addrspace(1)* undef
; 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(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i8
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[ARG2_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP9:%.*]] = load i32, i32 addrspace(4)* [[ARG3_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[ARG5_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG5_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP12:%.*]] = load i32, i32 addrspace(4)* [[ARG5_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 8, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[ARG6_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG6_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP15:%.*]] = load i32, i32 addrspace(4)* [[ARG6_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 8, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_I8_I8_I8_I8_I8_I8_I8_I8_KERNARG_SEGMENT]], i64 40
; MESA-NEXT:    [[ARG7_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG7_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP18:%.*]] = load i32, i32 addrspace(4)* [[ARG7_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 8, !invariant.load !0
; MESA-NEXT:    [[TMP19:%.*]] = lshr i32 [[TMP18]], 24
; MESA-NEXT:    [[TMP20:%.*]] = trunc i32 [[TMP19]] to i8
; MESA-NEXT:    store volatile i8 [[TMP2]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP5]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP8]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP11]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP14]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP17]], i8 addrspace(1)* undef
; MESA-NEXT:    store volatile i8 [[TMP20]], i8 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 %arg0, i8 addrspace(1)* undef
  store volatile i8 %arg1, i8 addrspace(1)* undef
  store volatile i8 %arg2, i8 addrspace(1)* undef
  store volatile i8 %arg3, i8 addrspace(1)* undef
  store volatile i8 %arg5, i8 addrspace(1)* undef
  store volatile i8 %arg6, i8 addrspace(1)* undef
  store volatile i8 %arg7, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; HSA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 16, !invariant.load !0
; 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]], half addrspace(1)* undef
; HSA-NEXT:    store volatile half [[ARG1_LOAD]], half addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_realign_f16_f16(
; MESA-NEXT:    [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP1:%.*]] = load i32, i32 addrspace(4)* [[ARG0_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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, i8 addrspace(4)* [[KERN_REALIGN_F16_F16_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN]] to i32 addrspace(4)*
; MESA-NEXT:    [[TMP3:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_ALIGN_DOWN_CAST]], align 4, !invariant.load !0
; 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]], half addrspace(1)* undef
; MESA-NEXT:    store volatile half [[ARG1_LOAD]], half addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile half %arg0, half addrspace(1)* undef
  store volatile half %arg1, half addrspace(1)* undef
  ret void
}

define amdgpu_kernel void @kern_global_ptr(i8 addrspace(1)* %ptr) #0 {
; HSA-LABEL: @kern_global_ptr(
; HSA-NEXT:    [[KERN_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_GLOBAL_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_global_ptr(
; MESA-NEXT:    [[KERN_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_GLOBAL_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 addrspace(1)* %ptr, i8 addrspace(1)* addrspace(1)* undef
  ret void
}

define amdgpu_kernel void @kern_global_ptr_dereferencable(i8 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(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_GLOBAL_PTR_DEREFERENCABLE_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0, !dereferenceable !1
; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_global_ptr_dereferencable(
; MESA-NEXT:    [[KERN_GLOBAL_PTR_DEREFERENCABLE_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_GLOBAL_PTR_DEREFERENCABLE_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0, !dereferenceable !1
; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 addrspace(1)* %ptr, i8 addrspace(1)* addrspace(1)* undef
  ret void
}

define amdgpu_kernel void @kern_global_ptr_dereferencable_or_null(i8 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(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_GLOBAL_PTR_DEREFERENCABLE_OR_NULL_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0, !dereferenceable_or_null !2
; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; 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(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_GLOBAL_PTR_DEREFERENCABLE_OR_NULL_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0, !dereferenceable_or_null !2
; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 addrspace(1)* %ptr, i8 addrspace(1)* addrspace(1)* undef
  ret void
}

define amdgpu_kernel void @kern_nonnull_global_ptr(i8 addrspace(1)* nonnull %ptr) #0 {
; HSA-LABEL: @kern_nonnull_global_ptr(
; HSA-NEXT:    [[KERN_NONNULL_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_NONNULL_GLOBAL_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0, !nonnull !0
; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_nonnull_global_ptr(
; MESA-NEXT:    [[KERN_NONNULL_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_NONNULL_GLOBAL_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0, !nonnull !0
; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 addrspace(1)* %ptr, i8 addrspace(1)* addrspace(1)* undef
  ret void
}

define amdgpu_kernel void @kern_align32_global_ptr(i8 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(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_ALIGN32_GLOBAL_PTR_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; HSA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0, !align !3
; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_align32_global_ptr(
; MESA-NEXT:    [[KERN_ALIGN32_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_ALIGN32_GLOBAL_PTR_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
; MESA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0, !align !3
; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 addrspace(1)* %ptr, i8 addrspace(1)* addrspace(1)* undef
  ret void
}

define amdgpu_kernel void @kern_noalias_global_ptr(i8 addrspace(1)* noalias %ptr) #0 {
; HSA-LABEL: @kern_noalias_global_ptr(
; HSA-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(8) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR:%.*]], i8 addrspace(1)* addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_noalias_global_ptr(
; MESA-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(44) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR:%.*]], i8 addrspace(1)* addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 addrspace(1)* %ptr, i8 addrspace(1)* addrspace(1)* undef
  ret void
}

define amdgpu_kernel void @kern_noalias_global_ptr_x2(i8 addrspace(1)* noalias %ptr0, i8 addrspace(1)* noalias %ptr1) #0 {
; HSA-LABEL: @kern_noalias_global_ptr_x2(
; HSA-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(16) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR0:%.*]], i8 addrspace(1)* addrspace(1)* undef
; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR1:%.*]], i8 addrspace(1)* addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @kern_noalias_global_ptr_x2(
; MESA-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(52) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR0:%.*]], i8 addrspace(1)* addrspace(1)* undef
; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR1:%.*]], i8 addrspace(1)* addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store volatile i8 addrspace(1)* %ptr0, i8 addrspace(1)* addrspace(1)* undef
  store volatile i8 addrspace(1)* %ptr1, i8 addrspace(1)* addrspace(1)* undef
  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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[STRUCT_I8_I8_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[IN_KERNARG_OFFSET]] to { i8, i8 } addrspace(4)*
; HSA-NEXT:    [[IN_LOAD:%.*]] = load { i8, i8 }, { i8, i8 } addrspace(4)* [[IN_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[ELT0:%.*]] = extractvalue { i8, i8 } [[IN_LOAD]], 0
; HSA-NEXT:    [[ELT1:%.*]] = extractvalue { i8, i8 } [[IN_LOAD]], 1
; HSA-NEXT:    store volatile i8 [[ELT0]], i8 addrspace(1)* null, align 4
; HSA-NEXT:    store volatile i8 [[ELT1]], i8 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[STRUCT_I8_I8_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[IN_KERNARG_OFFSET]] to { i8, i8 } addrspace(4)*
; MESA-NEXT:    [[IN_LOAD:%.*]] = load { i8, i8 }, { i8, i8 } addrspace(4)* [[IN_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[ELT0:%.*]] = extractvalue { i8, i8 } [[IN_LOAD]], 0
; MESA-NEXT:    [[ELT1:%.*]] = extractvalue { i8, i8 } [[IN_LOAD]], 1
; MESA-NEXT:    store volatile i8 [[ELT0]], i8 addrspace(1)* null, align 4
; MESA-NEXT:    store volatile i8 [[ELT1]], i8 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, i8 addrspace(1)* null, align 4
  store volatile i8 %elt1, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[STRUCT_I8_I16_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[IN_KERNARG_OFFSET]] to { i8, i16 } addrspace(4)*
; HSA-NEXT:    [[IN_LOAD:%.*]] = load { i8, i16 }, { i8, i16 } addrspace(4)* [[IN_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    [[ELT0:%.*]] = extractvalue { i8, i16 } [[IN_LOAD]], 0
; HSA-NEXT:    [[ELT1:%.*]] = extractvalue { i8, i16 } [[IN_LOAD]], 1
; HSA-NEXT:    store volatile i8 [[ELT0]], i8 addrspace(1)* null, align 4
; HSA-NEXT:    store volatile i16 [[ELT1]], i16 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[STRUCT_I8_I16_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[IN_KERNARG_OFFSET]] to { i8, i16 } addrspace(4)*
; MESA-NEXT:    [[IN_LOAD:%.*]] = load { i8, i16 }, { i8, i16 } addrspace(4)* [[IN_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    [[ELT0:%.*]] = extractvalue { i8, i16 } [[IN_LOAD]], 0
; MESA-NEXT:    [[ELT1:%.*]] = extractvalue { i8, i16 } [[IN_LOAD]], 1
; MESA-NEXT:    store volatile i8 [[ELT0]], i8 addrspace(1)* null, align 4
; MESA-NEXT:    store volatile i16 [[ELT1]], i16 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, i8 addrspace(1)* null, align 4
  store volatile i16 %elt1, i16 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[ARRAY_2XI8_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[IN_KERNARG_OFFSET]] to [2 x i8] addrspace(4)*
; HSA-NEXT:    [[IN_LOAD:%.*]] = load [2 x i8], [2 x i8] addrspace(4)* [[IN_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; 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]], i8 addrspace(1)* null, align 4
; HSA-NEXT:    store volatile i8 [[ELT1]], i8 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[ARRAY_2XI8_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[IN_KERNARG_OFFSET]] to [2 x i8] addrspace(4)*
; MESA-NEXT:    [[IN_LOAD:%.*]] = load [2 x i8], [2 x i8] addrspace(4)* [[IN_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; 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]], i8 addrspace(1)* null, align 4
; MESA-NEXT:    store volatile i8 [[ELT1]], i8 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, i8 addrspace(1)* null, align 4
  store volatile i8 %elt1, i8 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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[ARRAY_2XI1_ARG_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[IN_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[IN_KERNARG_OFFSET]] to [2 x i1] addrspace(4)*
; HSA-NEXT:    [[IN_LOAD:%.*]] = load [2 x i1], [2 x i1] addrspace(4)* [[IN_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; 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]], i1 addrspace(1)* null, align 4
; HSA-NEXT:    store volatile i1 [[ELT1]], i1 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(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[IN_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[ARRAY_2XI1_ARG_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[IN_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[IN_KERNARG_OFFSET]] to [2 x i1] addrspace(4)*
; MESA-NEXT:    [[IN_LOAD:%.*]] = load [2 x i1], [2 x i1] addrspace(4)* [[IN_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; 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]], i1 addrspace(1)* null, align 4
; MESA-NEXT:    store volatile i1 [[ELT1]], i1 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, i1 addrspace(1)* null, align 4
  store volatile i1 %elt1, i1 addrspace(1)* null, align 4
  ret void
}

define amdgpu_kernel void @only_empty_struct({} %empty) #0 {
; HSA-LABEL: @only_empty_struct(
; HSA-NEXT:    ret void
;
; MESA-LABEL: @only_empty_struct(
; MESA-NEXT:    [[ONLY_EMPTY_STRUCT_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(36) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-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(4) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[EMPTY_STRUCT_WITH_OTHER_KERNARG_SEGMENT]], i64 0
; HSA-NEXT:    [[ARG1_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET]] to i32 addrspace(4)*
; HSA-NEXT:    [[ARG1_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
; HSA-NEXT:    store i32 [[ARG1_LOAD]], i32 addrspace(1)* undef
; HSA-NEXT:    ret void
;
; MESA-LABEL: @empty_struct_with_other(
; MESA-NEXT:    [[EMPTY_STRUCT_WITH_OTHER_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(40) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[EMPTY_STRUCT_WITH_OTHER_KERNARG_SEGMENT]], i64 36
; MESA-NEXT:    [[ARG1_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[ARG1_KERNARG_OFFSET]] to i32 addrspace(4)*
; MESA-NEXT:    [[ARG1_LOAD:%.*]] = load i32, i32 addrspace(4)* [[ARG1_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
; MESA-NEXT:    store i32 [[ARG1_LOAD]], i32 addrspace(1)* undef
; MESA-NEXT:    ret void
;
  store i32 %arg1, i32 addrspace(1)* undef
  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" }

; GCN: 0 = !{}
; GCN: !1 = !{i64 42}
; GCN: !2 = !{i64 128}
; GCN: !3 = !{i64 1024}