Skip to content

Fix/dx12 shader f16 #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
dafioram merged 12 commits into from
Nov 5, 2026
Merged

Fix/dx12 shader f16 #4

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
e267553
Explain why DX12 FP16 is skipped
dafioram Nov 3, 2026
729658c
Add a DX12 FP16 experiment on WARP, with and without DXC
dafioram Nov 3, 2026
1ff4a5f
Warn that dxcompiler.dll can hide DX12 GPUs; show the compiler in --list
dafioram Nov 3, 2026
f4bd5ae
Name the K block behind a wrong matmul entry
dafioram Nov 4, 2026
cc1687c
Map which matmul tiles were written when the check fails
dafioram Nov 4, 2026
18d5b11
Create the matmul band offsets with their contents
dafioram Nov 4, 2026
b4678c3
Retry a failed matmul check as one dispatch over all rows
dafioram Nov 4, 2026
a8551d9
Use one matmul dispatch over all rows on software adapters
dafioram Nov 4, 2026
28830f1
Make each matmul band's bind group just before its dispatch
dafioram Nov 4, 2026
3b2d11c
Run the WARP CI matmul at N=2048 and ship README with the builds
dafioram Nov 4, 2026
145befb
Make the WARP CI gate fail unless the matmul ran at N=2048
dafioram Nov 5, 2026
d4cf924
Keep the requested matmul size on software adapters
dafioram Nov 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 10 additions & 2 deletions .github/workflows/build.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,8 +69,15 @@ jobs:
run: |
& .\target\release\gpu-workbench-cli.exe --list
if ($LASTEXITCODE -ne 0) { throw "--list failed with exit code $LASTEXITCODE" }
& .\target\release\gpu-workbench-cli.exe --cli --tests upload,readback,copy,bandwidth,matmul32,vram --transfer-mb 64 --matmul-n 256 --vram-mb 128 --vram-passes 1
if ($LASTEXITCODE -ne 0) { throw "CLI smoke test failed with exit code $LASTEXITCODE" }
$out = & .\target\release\gpu-workbench-cli.exe --cli --tests upload,readback,copy,bandwidth,matmul32,vram --transfer-mb 64 --matmul-n 2048 --vram-mb 128 --vram-passes 1 2>&1
$exit = $LASTEXITCODE
$out | ForEach-Object { "$_" }
if ($exit -ne 0) { throw "CLI smoke test failed with exit code $exit" }
# The planner shrinks N when its probe is slow, and the smaller run still
# prints PASS. The gate is only meaningful at N=2048, so check for that.
$text = $out | Out-String
if ($text -match 'auto-reduced') { throw "matmul was auto-reduced from N=2048; the gate did not test N=2048" }
if ($text -notmatch '\[PASS\][^\r\n]*N=2048,') { throw "no PASS line for the matmul at N=2048" }
# Refusing bad input must also exit non-zero from PowerShell.
- name: Refused options exit 2 (Windows)
if: runner.os == 'Windows'
Expand All @@ -93,3 +100,4 @@ jobs:
path: |
${{ matrix.gui }}
${{ matrix.cli }}
README.md
5 changes: 5 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -230,6 +230,11 @@ loaded. In PowerShell:
$env:WGPU_DX12_COMPILER = "dxc"
```

**Caution:** while `dxcompiler.dll` is next to the program, a GPU whose driver
does not support shader model 6.0 is hidden from the DX12 list (wgpu drops it
rather than fall back to FXC). Delete the DLLs to see that GPU again; it then
uses FXC. `gpu-workbench-cli --list` prints which compiler is in use.

## Log files

Every session writes a timestamped log: GPU and driver, sensor support,
Expand Down
Loading
Loading