170c0dac44
Strengthen out-of-bounds guarantees for buffer accesses by disallowing buffer accesses with alignment lower than natural alignment. This is needed to specifically address the edge case where an access starts out-of-bounds and then enters in-bounds, as the hardware would treat the entire access as being out-of-bounds. This is normally not needed for most users, but at least one graphics device extension (VK_EXT_robustness2) has very strict requirements - in-bounds accesses must return correct value, and out-of-bounds accesses must return zero. The direct consequence of the patch is that a buffer access at negative address is not merged by load-store-vectorizer with one at a positive address, which fixes a CTS test. Targets that do not care about the new behavior are advised to use the new target feature relaxed-buffer-oob-mode that maintains the state from before the patch.
32 lines
1.2 KiB
LLVM
32 lines
1.2 KiB
LLVM
; RUN: opt -S -mtriple=amdgcn-- -passes=load-store-vectorizer < %s | FileCheck -check-prefix=OPT %s
|
|
|
|
; OPT-LABEL: @buffer_fat_ptrs(
|
|
define void @buffer_fat_ptrs(ptr addrspace(7) %out) {
|
|
entry:
|
|
%a1 = getelementptr i32, ptr addrspace(7) %out, i32 1
|
|
%a2 = getelementptr i32, ptr addrspace(7) %out, i32 2
|
|
%a3 = getelementptr i32, ptr addrspace(7) %out, i32 3
|
|
|
|
; OPT: store <4 x i32> <i32 0, i32 1, i32 2, i32 3>, ptr addrspace(7) %out, align 16
|
|
store i32 0, ptr addrspace(7) %out, align 16
|
|
store i32 1, ptr addrspace(7) %a1, align 4
|
|
store i32 2, ptr addrspace(7) %a2, align 8
|
|
store i32 3, ptr addrspace(7) %a3, align 4
|
|
ret void
|
|
}
|
|
|
|
; OPT-LABEL: @buffer_strided_ptrs(
|
|
define void @buffer_strided_ptrs(ptr addrspace(9) %out) {
|
|
entry:
|
|
%a1 = getelementptr i32, ptr addrspace(9) %out, i32 1
|
|
%a2 = getelementptr i32, ptr addrspace(9) %out, i32 2
|
|
%a3 = getelementptr i32, ptr addrspace(9) %out, i32 3
|
|
|
|
; OPT: store <4 x i32> <i32 0, i32 1, i32 2, i32 3>, ptr addrspace(9) %out, align 16
|
|
store i32 0, ptr addrspace(9) %out, align 16
|
|
store i32 1, ptr addrspace(9) %a1, align 4
|
|
store i32 2, ptr addrspace(9) %a2, align 8
|
|
store i32 3, ptr addrspace(9) %a3, align 4
|
|
ret void
|
|
}
|