[NFC][mlir][linalg] Make conv_3d_ncdhw_fcdhw consistent with 2D variant #129547
Conversation
nirvedhmeshram
requested review from
dcaballe,
nicolasvasilache and
rengolin
as code owners
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-mlir-linalg Author: Nirvedh Meshram (nirvedhmeshram) ChangesOther convolutions such as conv_2d_nchw_fchw have a output affine map with no permutations and the input and the filter map are adjusted accordingly. This makes conv_3d_ncdhw_fcdhw a stand out so this PR changes the affine map to be consistent with other variants. Full diff: https://github.com/llvm/llvm-project/pull/129547.diff 1 Files Affected:
diff --git a/mlir/include/mlir/Dialect/Linalg/IR/LinalgNamedStructuredOps.yaml b/mlir/include/mlir/Dialect/Linalg/IR/LinalgNamedStructuredOps.yaml
index 496a323249e85..2bae43f887b04 100644
--- a/mlir/include/mlir/Dialect/Linalg/IR/LinalgNamedStructuredOps.yaml
+++ b/mlir/include/mlir/Dialect/Linalg/IR/LinalgNamedStructuredOps.yaml
@@ -4072,12 +4072,12 @@ structured_op: !LinalgStructuredOpConfig
indexing_maps: !LinalgIndexingMapsConfig
static_indexing_maps:
- affine_map<(d0, d1, d2, d3, d4, d5, d6, d7, d8)[s0, s1, s2, s3, s4, s5, s6,
- s7, s8, s9, s10, s11, s12, s13, s14] -> (d0, d8, d1 * s3 + d5 * s5, d2 * s7
- + d6 * s9, d3 * s11 + d7 * s13)>
+ s7, s8, s9, s10, s11, s12, s13, s14] -> (d0, d8, d2 * s3 + d5 * s5, d3 * s7
+ + d6 * s9, d4 * s11 + d7 * s13)>
- affine_map<(d0, d1, d2, d3, d4, d5, d6, d7, d8)[s0, s1, s2, s3, s4, s5, s6,
- s7, s8, s9, s10, s11, s12, s13, s14] -> (d4, d8, d5, d6, d7)>
+ s7, s8, s9, s10, s11, s12, s13, s14] -> (d1, d8, d5, d6, d7)>
- affine_map<(d0, d1, d2, d3, d4, d5, d6, d7, d8)[s0, s1, s2, s3, s4, s5, s6,
- s7, s8, s9, s10, s11, s12, s13, s14] -> (d0, d4, d1, d2, d3)>
+ s7, s8, s9, s10, s11, s12, s13, s14] -> (d0, d1, d2, d3, d4)>
iterator_types:
- parallel
- parallel
|
|
So effectively, for (d0, d1, d2, d3, d4, d5, d6, d7, d8) the interpretation:
For 2d nchw_fchw convolution's interpretations is (N, F, H, W, C, Hf, Wf), per llvm-project/mlir/include/mlir/Dialect/Linalg/IR/LinalgNamedStructuredOps.yaml Lines 3163 to 3170 in 84a7697 To be aligned in both cases I assume you also need to make 3d conv's C dimension |
nirvedhmeshram
force-pushed
the
3d_conv_nfc
branch
from
84a7697 to
c92ba10
Compare
Good catch, fixed that as well! |
…20150) The orginal generic im2col implementation assumed that the filter and output affine map would not have permutations. While this is true in most named ops, it is not necessarily required. See for example llvm/llvm-project#129547 where it was not the case for `conv_3d_ncdhw_fcdhw` . While in this case we just "normalized" the map upstream, this PR adds support for such maps directly. Fixes : #20139 --------- Signed-off-by: Nirvedh Meshram <[email protected]>
|
When I update |
#129547 changed the IR directly without updating the auto generate script. Signed-off-by: Nirvedh <[email protected]>
… (#133927) llvm/llvm-project#129547 changed the IR directly without updating the auto generate script. Signed-off-by: Nirvedh <[email protected]>
llvm/llvm-project#129547 changed the IR directly without updating the auto generate script. Signed-off-by: Nirvedh <[email protected]>
llvm/llvm-project#129547 changed the IR directly without updating the auto generate script. Signed-off-by: Nirvedh <[email protected]>
llvm/llvm-project#129547 changed the IR directly without updating the auto generate script. Signed-off-by: Nirvedh <[email protected]>
llvm/llvm-project#129547 changed the IR directly without updating the auto generate script. Signed-off-by: Nirvedh <[email protected]>
4 participants
Other convolutions such as conv_2d_nchw_fchw have a output affine map with no permutations and the input and the filter map are adjusted accordingly. This makes conv_3d_ncdhw_fcdhw a stand out so this PR changes the affine map to be consistent with other variants.