[TritonIntelGPU] Add Subgroup2DBlockLoadOpConversion for ttig.2d_block_load#6804
Open
whitneywhtsang wants to merge 4 commits intomainfrom
Open
[TritonIntelGPU] Add Subgroup2DBlockLoadOpConversion for ttig.2d_block_load#6804whitneywhtsang wants to merge 4 commits intomainfrom
whitneywhtsang wants to merge 4 commits intomainfrom
Conversation
whitneywhtsang
force-pushed
the
whitneywhtsang/Subgroup2DBlockLoadOpConversion
branch
from
4ee80de to
7fb3426
Compare
whitneywhtsang
force-pushed
the
whitneywhtsang/LowerTo2DBlockLoad_descriptor
branch
from
9e3cbf6 to
5a34d54
Compare
whitneywhtsang
force-pushed
the
whitneywhtsang/Subgroup2DBlockLoadOpConversion
branch
from
7fb3426 to
a2b36e6
Compare
whitneywhtsang
force-pushed
the
whitneywhtsang/LowerTo2DBlockLoad_descriptor
branch
2 times, most recently
from
81877c2 to
f75e245
Compare
whitneywhtsang
force-pushed
the
whitneywhtsang/Subgroup2DBlockLoadOpConversion
branch
from
a2b36e6 to
356731b
Compare
Base automatically changed from
whitneywhtsang/LowerTo2DBlockLoad_descriptor
to
main
May 1, 2026 22:58
whitneywhtsang
force-pushed
the
whitneywhtsang/Subgroup2DBlockLoadOpConversion
branch
from
356731b to
b904a45
Compare
Add rank-reducing (3D descriptor → 2D result) and same-rank batch (3D descriptor → 3D result) test cases to descriptor-load.mlir to cover the batch index folding into tt.addptr. Signed-off-by: Whitney Tsang <whitney.tsang@intel.com>
whitneywhtsang
force-pushed
the
whitneywhtsang/Subgroup2DBlockLoadOpConversion
branch
from
b904a45 to
653db35
Compare
…k_load Add LLVM lowering for ttig.2d_block_load → triton_gen.2Dblockload. The conversion mirrors DescriptorLoadOpToBlockIOConversion but reads decomposed surface parameters (width, height, pitch, offsets) directly from the op instead of unpacking a tensor descriptor struct. Verified to produce identical triton_gen.2Dblockload output as the old DescriptorLoadOpToBlockIOConversion for all 4 dot operand × memory layout combinations (dot A/B × row/column major) and NaN padding. Signed-off-by: Whitney Tsang <whitney.tsang@intel.com>
whitneywhtsang
force-pushed
the
whitneywhtsang/Subgroup2DBlockLoadOpConversion
branch
from
653db35 to
00ac019
Compare
Signed-off-by: Whitney Tsang <whitney.tsang@intel.com>
Descriptor loads eligible for 2D block I/O are now converted to ttig.2d_block_load by the LowerTo2DBlockLoad TTGIR pass and lowered by Subgroup2DBlockLoadOpConversion. Signed-off-by: Whitney Tsang <whitney.tsang@intel.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add LLVM lowering for ttig.2d_block_load → triton_gen.2Dblockload. The conversion mirrors DescriptorLoadOpToBlockIOConversion but reads decomposed surface parameters (width, height, pitch, offsets) directly from the op instead of unpacking a tensor descriptor struct.