…el fusion (#177065)

Add test_external_template_prologue_epilogue_fusion that exercises:
- Prologue fusion: sigmoid(b) fused into template as <LOAD_INPUT_B>
- Epilogue fusion: relu(...) * bias fused into template as <STORE_OUTPUT_0>
- Extra inputs: bias is read by the epilogue but is not among the
  template's original inputs, exercising kernel._extra_inputs

Uses a _MockExternalTemplateBuffer that subclasses TemplateBuffer and
creates an ExternalTritonTemplateKernel. The _render() method calls
kernel._setup_fusion_hooks() to set up all fusion hooks in one call,
then reads kernel._prologue_source_buffers and kernel._extra_store_targets
to build the template source with the appropriate placeholders.

Pull Request resolved: #177065
Approved by: https://github.com/jansel
ghstack dependencies: #177492

…tput templates (#177597)

TemplateBuffer subclasses with MultiOutputLayout (e.g. Helion kernels)
don't have a single dtype. Add an explicit error in TemplateBuffer.dtype
for this case, and guard the scheduler's low-precision heuristic with
is_multi_outputs_template() so it skips the check rather than crashing.

Pull Request resolved: #177597
Approved by: https://github.com/shunting314
ghstack dependencies: #177492, #177065

…onTemplateKernel (pytorch#177492)

Add fusion methods to ExternalTritonTemplateKernel:
- _compute_fusion_metadata(): determines eligible epilogues/prologues,
  builds epilogue specs, and computes prologue sources before render()
- _setup_fusion_hooks(): sets up epilogue/prologue render hooks and
  marks prologue buffers during render() (inside V.kernel context)
- _find_eligible_epilogues(): computes fusion eligibility and registers
  extra inputs needed by fused epilogues
- _setup_epilogue_hook(): creates independent subgraph with contiguous
  index state for epilogue codegen
- _setup_prologue_hook(): creates subgraph with capture store handler
  for prologue codegen, with variable renaming to avoid collisions

Differential Revision: [D96779294](https://our.internmc.facebook.com/intern/diff/D96779294)
Pull Request resolved: pytorch#177492
Approved by: https://github.com/jansel

…el fusion (pytorch#177065)

Add test_external_template_prologue_epilogue_fusion that exercises:
- Prologue fusion: sigmoid(b) fused into template as <LOAD_INPUT_B>
- Epilogue fusion: relu(...) * bias fused into template as <STORE_OUTPUT_0>
- Extra inputs: bias is read by the epilogue but is not among the
  template's original inputs, exercising kernel._extra_inputs

Uses a _MockExternalTemplateBuffer that subclasses TemplateBuffer and
creates an ExternalTritonTemplateKernel. The _render() method calls
kernel._setup_fusion_hooks() to set up all fusion hooks in one call,
then reads kernel._prologue_source_buffers and kernel._extra_store_targets
to build the template source with the appropriate placeholders.

Pull Request resolved: pytorch#177065
Approved by: https://github.com/jansel
ghstack dependencies: pytorch#177492

…tput templates (pytorch#177597)

TemplateBuffer subclasses with MultiOutputLayout (e.g. Helion kernels)
don't have a single dtype. Add an explicit error in TemplateBuffer.dtype
for this case, and guard the scheduler's low-precision heuristic with
is_multi_outputs_template() so it skips the check rather than crashing.

Pull Request resolved: pytorch#177597
Approved by: https://github.com/shunting314
ghstack dependencies: pytorch#177492, pytorch#177065

…onTemplateKernel (pytorch#177492)

Add fusion methods to ExternalTritonTemplateKernel:
- _compute_fusion_metadata(): determines eligible epilogues/prologues,
  builds epilogue specs, and computes prologue sources before render()
- _setup_fusion_hooks(): sets up epilogue/prologue render hooks and
  marks prologue buffers during render() (inside V.kernel context)
- _find_eligible_epilogues(): computes fusion eligibility and registers
  extra inputs needed by fused epilogues
- _setup_epilogue_hook(): creates independent subgraph with contiguous
  index state for epilogue codegen
- _setup_prologue_hook(): creates subgraph with capture store handler
  for prologue codegen, with variable renaming to avoid collisions

Differential Revision: [D96779294](https://our.internmc.facebook.com/intern/diff/D96779294)
Pull Request resolved: pytorch#177492
Approved by: https://github.com/jansel

…el fusion (pytorch#177065)

Add test_external_template_prologue_epilogue_fusion that exercises:
- Prologue fusion: sigmoid(b) fused into template as <LOAD_INPUT_B>
- Epilogue fusion: relu(...) * bias fused into template as <STORE_OUTPUT_0>
- Extra inputs: bias is read by the epilogue but is not among the
  template's original inputs, exercising kernel._extra_inputs

Uses a _MockExternalTemplateBuffer that subclasses TemplateBuffer and
creates an ExternalTritonTemplateKernel. The _render() method calls
kernel._setup_fusion_hooks() to set up all fusion hooks in one call,
then reads kernel._prologue_source_buffers and kernel._extra_store_targets
to build the template source with the appropriate placeholders.

Pull Request resolved: pytorch#177065
Approved by: https://github.com/jansel
ghstack dependencies: pytorch#177492

…tput templates (pytorch#177597)

TemplateBuffer subclasses with MultiOutputLayout (e.g. Helion kernels)
don't have a single dtype. Add an explicit error in TemplateBuffer.dtype
for this case, and guard the scheduler's low-precision heuristic with
is_multi_outputs_template() so it skips the check rather than crashing.

Pull Request resolved: pytorch#177597
Approved by: https://github.com/shunting314
ghstack dependencies: pytorch#177492, pytorch#177065