Draft
Conversation
Signed-off-by: Sanket Pandit <sanket.pandit@amd.com>
Signed-off-by: Sanket Pandit <sanket.pandit@amd.com>
Signed-off-by: Sanket Pandit <sanket.pandit@amd.com>
Signed-off-by: Sanket Pandit <sanket.pandit@amd.com>
Six infrastructure fixes needed for dynamic dims with the wave_asm backend: 1. scheduling/schedule.py: translate node_mapping keys back to original graph nodes after graph_copy in the dynamic pipelining path, fixing identity mismatch in _update_kernel_node_mapping (0/165 → 165/165). 2. wave_schedule_ops.py: use iterate's owning graph for subgraph reordering instead of hardcoding get_root_graph() (pipelined iterate lives inside a conditional subgraph with dynamic shapes). 3. unrolling.py: guard unroll count validation with is_literal() so symbolic counts don't raise TypeError. 4. emitter.py: handle Rational operands in Mod via the identity Mod(a/b,c)=Mod(a,b*c)/b; resolve terminal rationals with divsi. 5. read_write.py: initialize offset_th=None before masked code path (pre-existing bug only triggered by dynamic shapes). 6. host_codegen.py: resolve derived buffer shapes (K/2, K/32) for dynamic symbol recovery using infer_dim + sympy.solve, and evaluate dimension expressions via gen_sympy_index. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: Ivan Butygin <ivan.butygin@gmail.com>
Uses V_ASHRREV_I32 (arithmetic right shift) for signed division by power-of-2 constants, mirroring the existing arith.divui handler which uses V_LSHRREV_B32 (logical right shift). Required for dynamic M/N/K support where K/2 and K/32 appear as arith.divsi operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: Ivan Butygin <ivan.butygin@gmail.com>
When use_buffer_ops is enabled, avoid emitting vector<Nxindex> arith ops for bounds-check masks and OOB index selection. Instead: - _build_mask: new scalarize option builds per-element scalar cmpi and assembles the mask with vector.from_elements. - _create_vec_read_write: replace vector broadcast/addi/select with a scalar loop computing offset_th+i per element. - Enable use_buffer_ops in the dynamic-dims preshuffle-B test. This eliminates all vector<16xindex> ops from the dynamic-dims MLIR, which the WaveASM backend cannot translate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: Ivan Butygin <ivan.butygin@gmail.com>
- Add handleVectorFromElements: packs sub-dword scalars into VGPRs using V_LSHL_OR_B32 chains, combines DWORDs with PackOp. - Fix handleMemRefLoad to emit buffer_load_ubyte / ds_read_u8 for i8 element types instead of unconditionally emitting dword loads. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: Ivan Butygin <ivan.butygin@gmail.com>
Signed-off-by: Ivan Butygin <ivan.butygin@gmail.com>
Signed-off-by: Ivan Butygin <ivan.butygin@gmail.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
2 participants
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.
No description provided.