This is an archive of the discontinued LLVM Phabricator instance.

Differential D108005

[mlir][tosa] Fixed depthwise conv parallel/reduction indices order
ClosedPublic

Authored by rsuderman on Aug 12 2021, 3:50 PM.

Details

Reviewers
nicolasvasilache
NatashaKnk
antiagainst
Commits
rGf328f72e60a9: [mlir][tosa] Fixed depthwise conv parallel/reduction indices order
Summary

Reduction axis should come after all parallel axis to work with vectorization.

Diff Detail

Event Timeline

rsuderman created this revision.Aug 12 2021, 3:50 PM
Herald added subscribers: wrengr, Chia-hungDuan, shabalin and 21 others. · View Herald TranscriptAug 12 2021, 3:50 PM
rsuderman requested review of this revision.Aug 12 2021, 3:50 PM
Herald added a reviewer: nicolasvasilache. · View Herald TranscriptAug 12 2021, 3:50 PM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: stephenneuendorffer, nicolasvasilache. · View Herald Transcript
rsuderman added reviewers: NatashaKnk, antiagainst.Aug 12 2021, 3:50 PM
Harbormaster completed remote builds in B119356: Diff 366140.Aug 12 2021, 4:24 PM
NatashaKnk accepted this revision.Aug 16 2021, 1:44 PM
This revision is now accepted and ready to land.Aug 16 2021, 1:44 PM
Closed by commit rGf328f72e60a9: [mlir][tosa] Fixed depthwise conv parallel/reduction indices order (authored by rsuderman). · Explain WhyAug 16 2021, 2:07 PM
This revision was automatically updated to reflect the committed changes.
rsuderman added a commit: rGf328f72e60a9: [mlir][tosa] Fixed depthwise conv parallel/reduction indices order.

Revision Contents

PathSize
mlir/
include/
mlir/
Dialect/
Linalg/
IR/
20 lines
python/
mlir/
dialects/
linalg/
opdsl/
ops/
4 lines

Diff 366738

mlir/include/mlir/Dialect/Linalg/IR/LinalgNamedStructuredOps.yaml

Show First 20 Lines • Show All 943 Lines • ▼ Show 20 Lines - !LinalgOperandDefConfig
name: dilations name: dilations
usage: IndexAttribute usage: IndexAttribute
type_var: I64 type_var: I64
attribute_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, attribute_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11,
s12] -> (s11, s12)> s12] -> (s11, s12)>
indexing_maps: !LinalgIndexingMapsConfig indexing_maps: !LinalgIndexingMapsConfig
static_indexing_maps: static_indexing_maps:
- affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8, - affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8,
s9, s10, s11, s12] -> (d0, d1 * s9 + d3 * s11, d2 * s10 + d4 * s12, d5)> s9, s10, s11, s12] -> (d0, d1 * s9 + d5 * s11, d2 * s10 + d6 * s12, d3)>
- affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8, - affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8,
s9, s10, s11, s12] -> (d3, d4, d5, d6)> s9, s10, s11, s12] -> (d5, d6, d3, d4)>
- affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8, - affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8,
s9, s10, s11, s12] -> (d0, d1, d2, d5, d6)> s9, s10, s11, s12] -> (d0, d1, d2, d3, d4)>
iterator_types: iterator_types:
- parallel - parallel
- parallel - parallel
- parallel - parallel
- reduction
- reduction
- parallel - parallel
- parallel - parallel
- reduction
- reduction
assignments: assignments:
- !ScalarAssign - !ScalarAssign
arg: O arg: O
value: !ScalarExpression value: !ScalarExpression
scalar_apply: scalar_apply:
fn_name: add fn_name: add
operands: operands:
- !ScalarExpression - !ScalarExpression
▲ Show 20 Lines • Show All 61 Lines • ▼ Show 20 Lines - !LinalgOperandDefConfig
name: dilations name: dilations
usage: IndexAttribute usage: IndexAttribute
type_var: I64 type_var: I64
attribute_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, attribute_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11,
s12] -> (s11, s12)> s12] -> (s11, s12)>
indexing_maps: !LinalgIndexingMapsConfig indexing_maps: !LinalgIndexingMapsConfig
static_indexing_maps: static_indexing_maps:
- affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8, - affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8,
s9, s10, s11, s12] -> (d0, d1 * s9 + d3 * s11, d2 * s10 + d4 * s12, d5)> s9, s10, s11, s12] -> (d0, d1 * s9 + d5 * s11, d2 * s10 + d6 * s12, d3)>
- affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8, - affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8,
s9, s10, s11, s12] -> (d3, d4, d5, d6)> s9, s10, s11, s12] -> (d5, d6, d3, d4)>
- affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8, - affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8,
s9, s10, s11, s12] -> ()> s9, s10, s11, s12] -> ()>
- affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8, - affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8,
s9, s10, s11, s12] -> ()> s9, s10, s11, s12] -> ()>
- affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8, - affine_map<(d0, d1, d2, d3, d4, d5, d6)[s0, s1, s2, s3, s4, s5, s6, s7, s8,
s9, s10, s11, s12] -> (d0, d1, d2, d5, d6)> s9, s10, s11, s12] -> (d0, d1, d2, d3, d4)>
iterator_types: iterator_types:
- parallel - parallel
- parallel - parallel
- parallel - parallel
- reduction
- reduction
- parallel - parallel
- parallel - parallel
- reduction
- reduction
assignments: assignments:
- !ScalarAssign - !ScalarAssign
arg: O arg: O
value: !ScalarExpression value: !ScalarExpression
scalar_apply: scalar_apply:
fn_name: add fn_name: add
operands: operands:
- !ScalarExpression - !ScalarExpression
▲ Show 20 Lines • Show All 752 Lines • Show Last 20 Lines

mlir/python/mlir/dialects/linalg/opdsl/ops/core_named_ops.py

Show First 20 Lines • Show All 203 Lines • ▼ Show 20 Linesdef depthwise_conv2D_nhwc(
O=TensorDef(U, S.N, S.OH, S.OW, S.IC, S.CM, output=True), O=TensorDef(U, S.N, S.OH, S.OW, S.IC, S.CM, output=True),
strides=AttributeDef(S.SH, S.SW), strides=AttributeDef(S.SH, S.SW),
dilations=AttributeDef(S.DH, S.DW)): dilations=AttributeDef(S.DH, S.DW)):
"""Performs depth-wise 2-D convolution. """Performs depth-wise 2-D convolution.
Numeric casting is performed on the operands to the inner multiply, promoting Numeric casting is performed on the operands to the inner multiply, promoting
them to the same data type as the accumulator/output. them to the same data type as the accumulator/output.
""" """
domain(D.n, D.oh, D.ow, D.kh, D.kw, D.ic, D.cm) domain(D.n, D.oh, D.ow, D.ic, D.cm, D.kh, D.kw)
O[D.n, D.oh, D.ow, D.ic, D.cm] += cast( O[D.n, D.oh, D.ow, D.ic, D.cm] += cast(
U, I[D.n, D.oh * S.SH + D.kh * S.DH, D.ow * S.SW + D.kw * S.DW, U, I[D.n, D.oh * S.SH + D.kh * S.DH, D.ow * S.SW + D.kw * S.DW,
D.ic]) * cast(U, K[D.kh, D.kw, D.ic, D.cm]) D.ic]) * cast(U, K[D.kh, D.kw, D.ic, D.cm])
@linalg_structured_op @linalg_structured_op
def depthwise_conv2D_nhwc_q( def depthwise_conv2D_nhwc_q(
I=TensorDef(T1, S.N, S.IH, S.IW, S.IC), I=TensorDef(T1, S.N, S.IH, S.IW, S.IC),
K=TensorDef(T2, S.KH, S.KW, S.IC, S.CM), K=TensorDef(T2, S.KH, S.KW, S.IC, S.CM),
IZp=ScalarDef(I32), IZp=ScalarDef(I32),
KZp=ScalarDef(I32), KZp=ScalarDef(I32),
O=TensorDef(U, S.N, S.OH, S.OW, S.IC, S.CM, output=True), O=TensorDef(U, S.N, S.OH, S.OW, S.IC, S.CM, output=True),
strides=AttributeDef(S.SH, S.SW), strides=AttributeDef(S.SH, S.SW),
dilations=AttributeDef(S.DH, S.DW)): dilations=AttributeDef(S.DH, S.DW)):
"""Performs depth-wise 2-D convolution. """Performs depth-wise 2-D convolution.
Numeric casting is performed on the operands to the inner multiply, promoting Numeric casting is performed on the operands to the inner multiply, promoting
them to the same data type as the accumulator/output. them to the same data type as the accumulator/output.
""" """
domain(D.n, D.oh, D.ow, D.kh, D.kw, D.ic, D.cm) domain(D.n, D.oh, D.ow, D.ic, D.cm, D.kh, D.kw)
O[D.n, D.oh, D.ow, D.ic, D.cm] += ( O[D.n, D.oh, D.ow, D.ic, D.cm] += (
(cast(U, I[D.n, D.oh * S.SH + D.kh * S.DH, D.ow * S.SW + D.kw * S.DW, (cast(U, I[D.n, D.oh * S.SH + D.kh * S.DH, D.ow * S.SW + D.kw * S.DW,
D.ic]) - cast(U, IZp)) * D.ic]) - cast(U, IZp)) *
(cast(U, K[D.kh, D.kw, D.ic, D.cm]) - cast(U, KZp))) (cast(U, K[D.kh, D.kw, D.ic, D.cm]) - cast(U, KZp)))
@linalg_structured_op @linalg_structured_op
def pooling_nhwc_sum( def pooling_nhwc_sum(
▲ Show 20 Lines • Show All 161 Lines • Show Last 20 Lines