Skip to content

[Draft][GPU] Add LDS DMA alignment constraints to tiling heuristics#23357

Draft
Yu-Zhewen wants to merge 1 commit intoiree-org:mainfrom
Yu-Zhewen:dma_alignment
Draft

[Draft][GPU] Add LDS DMA alignment constraints to tiling heuristics#23357
Yu-Zhewen wants to merge 1 commit intoiree-org:mainfrom
Yu-Zhewen:dma_alignment

Conversation

@Yu-Zhewen
Copy link
Contributor

No description provided.

Comment on lines +49 to +65
/// Returns true if the target supports global load DMA (LDS DMA) operations.
/// Only CDNA4+ (gfx950 and newer) architectures support this feature.
/// Excludes RDNA cards (gfx10xx, gfx11xx, gfx12xx) which have major version
/// >= 10.
static bool targetSupportsGlobalLoadDMA(IREE::GPU::TargetAttr target) {
StringRef targetArch = target.getArch();
auto maybeChipset = amdgpu::Chipset::parse(targetArch);
if (failed(maybeChipset)) {
return false;
}
// Only enable for CDNA4+ (gfx950+). Exclude RDNA cards (gfx10xx, gfx11xx,
// gfx12xx). CDNA cards have major version 9, RDNA cards have major version
// >= 10.
constexpr amdgpu::Chipset kGfx950{9, 5, 0};
return maybeChipset->majorVersion == 9 && *maybeChipset >= kGfx950;
}

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is taken from #23230

Signed-off-by: Yu-Zhewen <zhewenyu@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant