[CIR][CodeGen] Special treatment of 3-element extended vector load and store #674
Conversation
jopperm
left a comment
jopperm
left a comment
There was a problem hiding this comment.
Could you add a test (or file an issue) for arrays of 3-component vectors, please?
| // CIR-NEXT: cir.store %[[#RESULT]], %[[#PVECC]] : !cir.vector<!s32i x 3>, !cir.ptr<!cir.vector<!s32i x 3>> | ||
|
|
||
| // LLVM-NEXT: %[[#VECB:]] = load <2 x i32>, ptr %[[#PVECB]], align 8 | ||
| // LLVM-NEXT: %[[#VECC:]] = load <3 x i32>, ptr %[[#PVECC]], align 16 |
There was a problem hiding this comment.
This matches clang's codegen, but do we understand why the vector is not loaded as <4 x i32> here?
There was a problem hiding this comment.
Seems a mistake from upstream?
There was a problem hiding this comment.
Not blocking the review, but I agree with @jopperm, since you're going to be looking at OpenCL it might be good to try to understand why for your own knowledge of how vectors should play out in general.
bcardosolopes
left a comment
bcardosolopes
left a comment
There was a problem hiding this comment.
LGTM with few clarifying questions / comments for future PRs.
| // CIR-NEXT: cir.store %[[#RESULT]], %[[#PVECC]] : !cir.vector<!s32i x 3>, !cir.ptr<!cir.vector<!s32i x 3>> | ||
|
|
||
| // LLVM-NEXT: %[[#VECB:]] = load <2 x i32>, ptr %[[#PVECB]], align 8 | ||
| // LLVM-NEXT: %[[#VECC:]] = load <3 x i32>, ptr %[[#PVECC]], align 16 |
There was a problem hiding this comment.
Not blocking the review, but I agree with @jopperm, since you're going to be looking at OpenCL it might be good to try to understand why for your own knowledge of how vectors should play out in general.
3 participants
Continue the work of #613 .
Original CodeGen treat vec3 as vec4 to get aligned memory access. This PR enable these paths.