Repository navigation
perf: s390x float32 Max/Min (VFMAXSB, z14+) - #69
Merged
Merged
Conversation
…measured max32VXE/min32VXE: VFMAXSB/VFMINSB in mode 1 (Java: NaN-propagating, -0 below +0), four four-lane accumulators folded by an 8-byte VSLDB rotation, behind useVXE, from 16 elements up. Measured on a real IBM Z (abbench, 6 rounds, alternated): float32 Max -98.3% (2^14) / -97.2% (2^20) against v0.22.0; and, against v0.17.0, the float32 kernels that had only been tested there: Add -80/-85%, AddInto -92/-89%, Sqrt -82/-83%. float64 controls flat. A sabotaged kernel (one VFMINSB lane) fails the sequential-oracle test on the real machine and, in a first push of this PR, in CI's s390x qemu lane (TestExtrema32MatchSequential, same message). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tannevaled
force-pushed
the
s390x-extrema32
branch
from
October 26, 2026 12:48
6d95c49 to
385381d
Compare
tannevaled
marked this pull request as ready for review
October 26, 2026 12:48
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
s390x float32
Max/Min, and the first real-hardware measurements of the z14 float32 kernels.Kernel.
max32VXE/min32VXEuseVFMAXSB/VFMINSBin mode 1 (Java's Math.max/min), so a NaN operand gives NaN and −0 orders below +0. Four four-lane accumulators are folded by an 8-byteVSLDBrotation, then lane 1 onto lane 0. They run behinduseVXE(z14), from 16 elements up.Measured on a real IBM Z (LinuxONE,
abbench, 6 rounds, alternated, pinned core):MaxAddAddIntoSqrtThe float64 controls are flat in both runs. The z machine had been unreachable since 2026-10-25 because of a local SSH agent problem; see
docs/perf.md. So the v0.18.0/v0.19.0 float32 kernels had been tested there but never timed.Proof the tests can fail.
VFMINSBlane inmax32VXE. CI's s390x qemu lane failed exactly there (TestExtrema32MatchSequential: special Max32 n=32: -0, sequential 0). That lane runs VXE since ci: run the s390x qemu lane with VXE, and require it #68.NDARRAY_REQUIRE_VXE=1, and so does the root suite.The local coverage gate (
covblocks) reports 0 uncovered blocks.A minor release, v0.23.0.
🤖 Generated with Claude Code