95 lines
5.1 KiB
LLVM
95 lines
5.1 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -amdgpu-late-codegenprepare %s | FileCheck %s -check-prefix=GFX9
|
|
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -amdgpu-late-codegenprepare %s | FileCheck %s -check-prefix=GFX12
|
|
|
|
; Make sure we don't crash when trying to create a bitcast between
|
|
; address spaces
|
|
define amdgpu_kernel void @constant_from_offset_cast_generic_null() {
|
|
; GFX9-LABEL: @constant_from_offset_cast_generic_null(
|
|
; GFX9-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(4) getelementptr (i8, ptr addrspace(4) addrspacecast (ptr null to ptr addrspace(4)), i64 4), align 4
|
|
; GFX9-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 16
|
|
; GFX9-NEXT: [[TMP3:%.*]] = trunc i32 [[TMP2]] to i8
|
|
; GFX9-NEXT: store i8 [[TMP3]], ptr addrspace(1) undef, align 1
|
|
; GFX9-NEXT: ret void
|
|
;
|
|
; GFX12-LABEL: @constant_from_offset_cast_generic_null(
|
|
; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr null to ptr addrspace(4)), i64 6), align 1
|
|
; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1
|
|
; GFX12-NEXT: ret void
|
|
;
|
|
%load = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr null to ptr addrspace(4)), i64 6), align 1
|
|
store i8 %load, ptr addrspace(1) undef
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @constant_from_offset_cast_global_null() {
|
|
; GFX9-LABEL: @constant_from_offset_cast_global_null(
|
|
; GFX9-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(4) getelementptr (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) null to ptr addrspace(4)), i64 4), align 4
|
|
; GFX9-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 16
|
|
; GFX9-NEXT: [[TMP3:%.*]] = trunc i32 [[TMP2]] to i8
|
|
; GFX9-NEXT: store i8 [[TMP3]], ptr addrspace(1) undef, align 1
|
|
; GFX9-NEXT: ret void
|
|
;
|
|
; GFX12-LABEL: @constant_from_offset_cast_global_null(
|
|
; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) null to ptr addrspace(4)), i64 6), align 1
|
|
; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1
|
|
; GFX12-NEXT: ret void
|
|
;
|
|
%load = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) null to ptr addrspace(4)), i64 6), align 1
|
|
store i8 %load, ptr addrspace(1) undef
|
|
ret void
|
|
}
|
|
|
|
@gv = unnamed_addr addrspace(1) global [64 x i8] undef, align 4
|
|
|
|
define amdgpu_kernel void @constant_from_offset_cast_global_gv() {
|
|
; GFX9-LABEL: @constant_from_offset_cast_global_gv(
|
|
; GFX9-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(4) getelementptr (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) @gv to ptr addrspace(4)), i64 4), align 4
|
|
; GFX9-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 16
|
|
; GFX9-NEXT: [[TMP3:%.*]] = trunc i32 [[TMP2]] to i8
|
|
; GFX9-NEXT: store i8 [[TMP3]], ptr addrspace(1) undef, align 1
|
|
; GFX9-NEXT: ret void
|
|
;
|
|
; GFX12-LABEL: @constant_from_offset_cast_global_gv(
|
|
; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) @gv to ptr addrspace(4)), i64 6), align 1
|
|
; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1
|
|
; GFX12-NEXT: ret void
|
|
;
|
|
%load = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) @gv to ptr addrspace(4)), i64 6), align 1
|
|
store i8 %load, ptr addrspace(1) undef
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @constant_from_offset_cast_generic_inttoptr() {
|
|
; GFX9-LABEL: @constant_from_offset_cast_generic_inttoptr(
|
|
; GFX9-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(4) getelementptr (i8, ptr addrspace(4) addrspacecast (ptr inttoptr (i64 128 to ptr) to ptr addrspace(4)), i64 4), align 4
|
|
; GFX9-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 16
|
|
; GFX9-NEXT: [[TMP3:%.*]] = trunc i32 [[TMP2]] to i8
|
|
; GFX9-NEXT: store i8 [[TMP3]], ptr addrspace(1) undef, align 1
|
|
; GFX9-NEXT: ret void
|
|
;
|
|
; GFX12-LABEL: @constant_from_offset_cast_generic_inttoptr(
|
|
; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr inttoptr (i64 128 to ptr) to ptr addrspace(4)), i64 6), align 1
|
|
; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1
|
|
; GFX12-NEXT: ret void
|
|
;
|
|
%load = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr inttoptr (i64 128 to ptr) to ptr addrspace(4)), i64 6), align 1
|
|
store i8 %load, ptr addrspace(1) undef
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @constant_from_inttoptr() {
|
|
; GFX9-LABEL: @constant_from_inttoptr(
|
|
; GFX9-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) inttoptr (i64 128 to ptr addrspace(4)), align 4
|
|
; GFX9-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1
|
|
; GFX9-NEXT: ret void
|
|
;
|
|
; GFX12-LABEL: @constant_from_inttoptr(
|
|
; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) inttoptr (i64 128 to ptr addrspace(4)), align 1
|
|
; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1
|
|
; GFX12-NEXT: ret void
|
|
;
|
|
%load = load i8, ptr addrspace(4) inttoptr (i64 128 to ptr addrspace(4)), align 1
|
|
store i8 %load, ptr addrspace(1) undef
|
|
ret void
|
|
}
|