diff options
| author | Christian Krinitsin <mail@krinitsin.com> | 2025-07-17 09:10:43 +0200 |
|---|---|---|
| committer | Christian Krinitsin <mail@krinitsin.com> | 2025-07-17 09:10:43 +0200 |
| commit | f2ec263023649e596c5076df32c2d328bc9393d2 (patch) | |
| tree | 5dd86caab46e552bd2e62bf9c4fb1a7504a44db4 /results/scraper/fex/3806 | |
| parent | 63d2e9d409831aa8582787234cae4741847504b7 (diff) | |
| download | qemu-analysis-main.tar.gz qemu-analysis-main.zip | |
Diffstat (limited to 'results/scraper/fex/3806')
| -rw-r--r-- | results/scraper/fex/3806 | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/results/scraper/fex/3806 b/results/scraper/fex/3806 new file mode 100644 index 000000000..c227868cd --- /dev/null +++ b/results/scraper/fex/3806 @@ -0,0 +1,7 @@ +AVX128: Extend 32-bit indices in the case of `VPGATHERDQ ymm1, vm32x, ymm2` +For the particular code path that `VPGATHERDQ ymm1, vm32x, ymm2` hits, we can sign-extend the indices provided in vm32x to be 64-bit indices. +This only really works because we split the IR operation in to two halves anyway, so 4 32-bit indices in vm32x loading 4 64-bit indices in ymm1 works out to two SVE gather loads using 64-bit indices quite well. + +We can also then use the scaling optimization that #3805 proposes since the 32-bit signed indices are guaranteed to overflow correctly in this case. + + |