AI: [G3] dequantize_per_tensor asym SIMD: integer zero-point subtract#20703
Open
zonglinpeng wants to merge 2 commits into
Open
AI: [G3] dequantize_per_tensor asym SIMD: integer zero-point subtract#20703zonglinpeng wants to merge 2 commits into
zonglinpeng wants to merge 2 commits into
Conversation
…20499) Summary: Recreates the optimized Fusion-G3 dequantize from D108798741, but instead of shipping it as a separate devmate kernel wired in through `operator_fallback.bzl`, it places the PDX SIMD fast path directly into the existing executorch operator `dequantize_per_tensor_out` in `executorch/backends/cadence/fusion_g3/operators/op_dequantize.cpp` (per-tensor function only; `per_channel`/`tensor`/`tensor_args` variants are untouched). When the input and output buffers are 16-byte aligned (`dequant_simd_aligned`), the per-tensor path runs an inline PDX SIMD loop (`xb_vecMxf32`/`xb_vecMx32`/`PDX_MUL_MXF32`); otherwise it falls back to the NNLib path (`xa_nn_elm_dequantize_*`). The result is numerically identical to the original op — the same float-domain affine `(x - zero_point) * scale`. This intentionally does NOT include the mvartanian integer-subtract change (D109458111, `PDX_SUB_MX32`); it uses the float-domain asymmetric path from D108798741 as requested. The macro fast paths (`ASYM_DEQUANTIZE_IMPL_CHANNEL`/`SYM_DEQUANTIZE_IMPL_CHANNEL`) get the `static_cast<CTYPE_OUT>((x - zp) * scale)` parenthesization required to build clean under the G3 `dev` mode's `-Werror,-Wdouble-promotion`. For A/B measurement this also adds `op_dequantize_baseline.cpp` under the Jarvis operator test dir: a benchmark-only snapshot of the ORIGINAL executorch op (pre-SIMD, with only the `-Wdouble-promotion` fix). It defines `impl::G3::native::dequantize_per_tensor_out`, so the shared benchmark source from D109441948 is linked into two binaries — `_optimized` (against the real executorch op) and `_stock` (against the snapshot) — and compared on the cycle-accurate G3 ISS. `operators_header` visibility is extended to the Jarvis test package so the snapshot can include `operators.h`. Reviewed By: mvartani-meta Differential Revision: D109500113
Summary: Optimizes the per-tensor `dequantize_per_tensor_out` SIMD fast path on Fusion-G3 for int8 / uint8 / int16 / uint16 -> float32. What is optimized: the zero-point subtract is kept in the int32 domain (`PDX_SUB_MX32`) and the int->float conversion is folded into the multiply by feeding the widened int32 codes straight into the mixed-type `PDX_MUL_MXF32(int32, scale)` (the idiom the vendor NNLib dequantize kernel uses). Per element the symmetric path is then a single fused multiply, and the asymmetric path is one integer subtract plus one fused multiply, with no separate int->float convert instruction on either path. Bit-identical to the previous code for 8/16-bit inputs, since `q - zp` is exact in int32 and converts exactly to float, matching `float(q) - float(zp)`. What is improved from 00b95cf2b4: that revision moved the subtract into the integer domain but kept an explicit `(xb_vecMxf32)` convert, so the asymmetric FP pipe still did convert + multiply (a `sub -> convert -> mul` dependency chain) and the symmetric path was left untouched (also convert + multiply). This diff removes that convert on both paths via the fused mixed-type multiply, shortening the asymmetric chain to `sub -> fused-mul` and the symmetric chain to a single fused multiply. That clears the residual large-tensor asymmetric regression 00b95cf2b4 still had -- e.g. `objv1_1x400x400x1_u8` (160000 elems) goes 0.75x -> 1.00x vs stock and `dpev26_4x60x4x4_u16` goes 0.89x -> 1.17x -- while keeping every symmetric case at parity or better, and the object file is ~144 B smaller. Differential Revision: D110529287
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20703
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit f81044d with merge base 8965e51 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
Contributor
|
@zonglinpeng has exported this pull request. If you are a Meta employee, you can view the originating Diff in D110529287. |
This PR needs a
|
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.
Summary:
Optimizes the per-tensor
dequantize_per_tensor_outSIMD fast path on Fusion-G3 for int8 / uint8 / int16 / uint16 -> float32.What is optimized: the zero-point subtract is kept in the int32 domain (
PDX_SUB_MX32) and the int->float conversion is folded into the multiply by feeding the widened int32 codes straight into the mixed-typePDX_MUL_MXF32(int32, scale)(the idiom the vendor NNLib dequantize kernel uses). Per element the symmetric path is then a single fused multiply, and the asymmetric path is one integer subtract plus one fused multiply, with no separate int->float convert instruction on either path. Bit-identical to the previous code for 8/16-bit inputs, sinceq - zpis exact in int32 and converts exactly to float, matchingfloat(q) - float(zp).What is improved from 00b95cf2b4: that revision moved the subtract into the integer domain but kept an explicit
(xb_vecMxf32)convert, so the asymmetric FP pipe still did convert + multiply (asub -> convert -> muldependency chain) and the symmetric path was left untouched (also convert + multiply). This diff removes that convert on both paths via the fused mixed-type multiply, shortening the asymmetric chain tosub -> fused-muland the symmetric chain to a single fused multiply. That clears the residual large-tensor asymmetric regression 00b95cf2b4 still had -- e.g.objv1_1x400x400x1_u8(160000 elems) goes 0.75x -> 1.00x vs stock anddpev26_4x60x4x4_u16goes 0.89x -> 1.17x -- while keeping every symmetric case at parity or better, and the object file is ~144 B smaller.Differential Revision: D110529287