| commit | c4b591a10f2f604291da14aea7983a5fa7837ce8 | |
| author | Luke Lau <luke@igalia.com> | |
| Mon, 15 Jan 2024 08:52:20 +0000 (15 15:52 +0700) | ||
| committer | GitHub <noreply@github.com> | |
| Mon, 15 Jan 2024 08:52:20 +0000 (15 15:52 +0700) | ||
| tree | 0b1ef4e332db0fb29aca3fd0fc143a403a71019f | treesnapshot (tar.gz zip) |
| parent | 8cac995eadebbf23b10f8cebc9a42adcf91cc26b | commitdiff |
[Flang][RISCV] Set vscale_range based off zvl*b (#77277)
This patch implements the logic (for now, copied from
RISCVTargetInfo::getVScaleRange) so that we can compute the vscale_range
based off of the zvl*b extension, e.g. using an arch with zvl256b now
implies vscale_range(2,1024).
It's worth noting that we don't have to exactly copy the behaviour of
clang with regards to how it interacts with the
-mvscale-min/-mvscale-max flags, but changing it can be left to a future
patch.
This also adds a guard for +sve so that we only check for it on aarch64,
which was the behaviour prior to 898db1136e679
This patch implements the logic (for now, copied from
RISCVTargetInfo::getVScaleRange) so that we can compute the vscale_range
based off of the zvl*b extension, e.g. using an arch with zvl256b now
implies vscale_range(2,1024).
It's worth noting that we don't have to exactly copy the behaviour of
clang with regards to how it interacts with the
-mvscale-min/-mvscale-max flags, but changing it can be left to a future
patch.
This also adds a guard for +sve so that we only check for it on aarch64,
which was the behaviour prior to 898db1136e679
| flang/lib/Frontend/FrontendActions.cpp | diffblobblamehistory | |
| flang/test/Lower/RISCV/riscv-vector-bits-vscale-range.f90 | diffblobblamehistory |