Add toggle for wavefront size to cross compilation options

[bdevorem]

Motivation

For cross compiling, we want to be able to toggle the wavefront size for architectures instead of just letting the decision be made strictly on whether or not the arch is gfx11 or gfx12.

Technical Details

This PR adds wavefront size as a toggle, which can be directly set via JSON object with the driver --gpu-arch-params option, or directly with a CLI flag --gpu-wavefront-size, or the various language APIs:

1. --gpu-arch-params JSON (primary, matches other toggles)

migraphx-driver compile model.onnx --gpu-arch gfx1200 \
  --gpu-arch-params "{wavefront_size:64, num_cu:60}"

2. Dedicated CLI flag

migraphx-driver compile model.onnx --gpu-arch gfx1200 --gpu-wavefront-size 64

Same as --gpu-num-cus style. --gpu-arch-params overrides this if both are set (same precedence pattern as the other params).

3. C++ target API

auto t = migraphx::make_target("gpu", migraphx::value{
    {"gpu_arch", "gfx1200"},
    {"gpu_wavefront_size", 64},
});
p.compile(t);

4. C API

migraphx::target t("gpu", "{gpu_arch: gfx1200, gpu_wavefront_size: 64

5. Python API

t = migraphx.get_target("gpu", gpu_arch="gfx1200", gpu_wavefront_size=64)
p.compile(t)

Works automatically via reflect() on gpu::target

List was generated by cursor.

Changelog Category

Add a CHANGELOG.md entry for any option other than Not Applicable

• • Added: New functionality.
• • Changed: Changes to existing functionality.
• • Removed: Functionality or support that has been removed. (Compared to a previous release)
• • Optimized: Component performance that has been optimized or improved.
• • Resolved Issues: Known issues from a previous version that have been resolved.
• • Not Applicable: This PR is not to be included in the changelog.

Follow the LLVM AI Tool Use Policy for contributions using AI.

[Copilot]

Pull request overview

Adds an explicit wavefront-size override to MIGraphX GPU cross-compilation configuration, enabling users to force wave32/wave64 behavior during cross-compilation rather than relying only on architecture-based inference.

Changes:

• Plumbed a new gpu_wavefront_size target option through GPU cross-compile context creation and synthetic hipDeviceProp_t generation.
• Added --gpu-wavefront-size and wavefront_size support in --gpu-arch-params for migraphx-driver.
• Updated tests, developer documentation, and changelog to cover the new option.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/targets/gpu/include/migraphx/gpu/target.hpp	Adds gpu_wavefront_size to the GPU target’s reflected/serialized options.
src/targets/gpu/target.cpp	Passes gpu_wavefront_size into cross-compile context creation.
src/targets/gpu/include/migraphx/gpu/context.hpp	Extends cross-compile hip_device/context constructors to accept a wavefront-size override.
src/targets/gpu/include/migraphx/gpu/cross_compile_device.hpp	Extends the cross-compile device-props API with a wavefront_size parameter.
src/targets/gpu/cross_compile_device.cpp	Validates and applies the wavefront-size override when populating synthetic HIP device properties.
src/driver/main.cpp	Adds --gpu-wavefront-size and JSON key mapping to pass wavefront size into GPU cross-compilation options.
test/gpu/jit.cpp	Adds coverage for default vs overridden wavefront size in cross-compile device props.
test/gpu/target_serialize.cpp	Adds serialization round-trip coverage for gpu_wavefront_size.
docs/dev/cross_compilation.rst	Documents wavefront-size inference and explicit override mechanisms for cross-compilation.
CHANGELOG.md	Adds an entry describing the new cross-compile wavefront size configuration.

[gh-app-migraphx-bot-pr-write]

Test	Batch	New Rate (d85196)	Old Rate (55ca93)	Diff	Status
torchvision-resnet50	64	3,153.69	3,169.20	-0.49%	✅
torchvision-resnet50_fp16	64	6,662.68	6,721.88	-0.88%	✅
torchvision-densenet121	32	2,695.78	2,712.99	-0.63%	✅
torchvision-densenet121_fp16	32	4,552.33	4,578.90	-0.58%	✅
torchvision-inceptionv3	32	1,798.05	1,805.02	-0.39%	✅
torchvision-inceptionv3_fp16	32	2,844.22	2,855.31	-0.39%	✅
cadene-inceptionv4	16	821.14	824.13	-0.36%	✅
cadene-resnext64x4	16	783.01	785.46	-0.31%	✅
slim-mobilenet	64	8,367.70	8,444.46	-0.91%	✅
slim-nasnetalarge	64	229.27	229.50	-0.10%	✅
slim-resnet50v2	64	3,167.58	3,185.35	-0.56%	✅
bert-mrpc-onnx	8	1,171.17	1,173.56	-0.20%	✅
bert-mrpc-tf	1	480.61	484.20	-0.74%	✅
pytorch-examples-wlang-gru	1	475.17	472.27	0.61%	✅
pytorch-examples-wlang-lstm	1	386.47	386.76	-0.07%	✅
torchvision-resnet50_1	1	748.11	754.58	-0.86%	✅
cadene-dpn92_1	1	440.34	444.58	-0.95%	✅
cadene-resnext101_1	1	365.69	366.32	-0.17%	✅
onnx-taau-downsample	1	401.22	402.99	-0.44%	✅
dlrm-criteoterabyte	1	32.54	32.58	-0.12%	✅
dlrm-criteoterabyte_fp16	1	52.06	52.65	-1.12%	✅
agentmodel	1	8,733.40	7,784.58	12.19%	🔆
unet_fp16	2	57.08	57.48	-0.70%	✅
resnet50v1_fp16	1	955.02	932.89	2.37%	✅
resnet50v1_int8	1	939.03	934.74	0.46%	✅
bert_base_cased_fp16	64	1,098.61	1,105.09	-0.59%	✅
bert_large_uncased_fp16	32	345.48	347.51	-0.58%	✅
bert_large_fp16	1	205.36	205.61	-0.12%	✅
distilgpt2_fp16	16	2,087.51	2,104.89	-0.83%	✅
yolov5s	1	560.66	559.63	0.18%	✅
tinyllama	1	45.81	46.13	-0.70%	✅
vicuna-fastchat	1	44.07	44.16	-0.20%	✅
whisper-tiny-encoder	1	411.84	415.45	-0.87%	✅
whisper-tiny-decoder	1	408.80	412.20	-0.83%	✅
llama2_7b	1	20.85	21.00	-0.73%	✅
qwen1.5-7b	1	14.95	23.73	-37.00%	🔴
phi3-3.8b	1	26.67	26.83	-0.60%	✅
llama3-8b	1	14.04	21.87	-35.83%	🔴
whisper-large-encoder	1	10.16	10.23	-0.72%	✅
whisper-large-decoder	1	104.97	104.89	0.08%	✅
mistral-7b	1	6.73	23.89	-71.83%	🔴
FLUX.1-schnell	1	146.71	760.54	-80.71%	🔴

Regressions detected 🔴

[gh-app-migraphx-bot-pr-write]

Test	Status	Result
bert-mrpc-onnx	✅	PASSED: MIGraphX meets tolerance
bert-mrpc-tf	✅	PASSED: MIGraphX meets tolerance
pytorch-examples-wlang-gru	✅	PASSED: MIGraphX meets tolerance
pytorch-examples-wlang-lstm	✅	PASSED: MIGraphX meets tolerance
dlrm-criteoterabyte	✅	PASSED: MIGraphX meets tolerance
agentmodel	✅	PASSED: MIGraphX meets tolerance
unet	✅	PASSED: MIGraphX meets tolerance
resnet50v1	✅	PASSED: MIGraphX meets tolerance
bert_base_cased_fp16	✅	PASSED: MIGraphX meets tolerance
bert_large_uncased_fp16	🔴	FAILED: MIGraphX is not within tolerance - check verbose output
bert_large	✅	PASSED: MIGraphX meets tolerance
yolov5s	✅	PASSED: MIGraphX meets tolerance
tinyllama	✅	PASSED: MIGraphX meets tolerance
vicuna-fastchat	✅	PASSED: MIGraphX meets tolerance
whisper-tiny-encoder	✅	PASSED: MIGraphX meets tolerance
whisper-tiny-decoder	✅	PASSED: MIGraphX meets tolerance
distilgpt2_fp16	✅	PASSED: MIGraphX meets tolerance
llama2_7b	✅	PASSED: MIGraphX meets tolerance
qwen1.5-7b	✅	PASSED: MIGraphX meets tolerance
phi3-3.8b	✅	PASSED: MIGraphX meets tolerance
llama3-8b	✅	PASSED: MIGraphX meets tolerance
whisper-large-encoder	✅	PASSED: MIGraphX meets tolerance
whisper-large-decoder	✅	PASSED: MIGraphX meets tolerance
mistral-7b	✅	PASSED: MIGraphX meets tolerance
FLUX.1-schnell	✅	PASSED: MIGraphX meets tolerance

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #5065      +/-   ##
===========================================
+ Coverage    92.89%   92.90%   +0.01%     
===========================================
  Files          603      603              
  Lines        32448    32526      +78     
===========================================
+ Hits         30140    30217      +77     
- Misses        2308     2309       +1     

see 6 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.