…no epilogue_nodes (#136518)
The `template_buffer_has_other_users` function checks the case where there're epilogue nodes and the template output has users other than these epilogue nodes.  When there's no epilogue nodes, the function could return `False` directly.
Pull Request resolved: #136518
Approved by: https://github.com/leslie-fang-intel, https://github.com/jgong5
ghstack dependencies: #136418

…ytorch#136418)
Fixes the compilation error of max-autotune for `maml_omniglot` (AMP and FP32) and `soft_actor_critic` (AMP) in Torchbench for single-thread dynamic shapes case:
```
/tmp/torchinductor_user/uv/cuvq6wenwp7us423onuvntkfx4cspmagha5beiknob7tiebzhupa.cpp: In function ‘void kernel(const bfloat16*, const bfloat16*, const bfloat16*, bfloat16*, int64_t)’:
/tmp/torchinductor_user/uv/cuvq6wenwp7us423onuvntkfx4cspmagha5beiknob7tiebzhupa.cpp:279:41: error: the value of ‘Mr_blocks’ is not usable in a constant expression
  279 |         constexpr int64_t m_block_end = Mr_blocks;
      |                                         ^~~~~~~~~
/tmp/torchinductor_user/uv/cuvq6wenwp7us423onuvntkfx4cspmagha5beiknob7tiebzhupa.cpp:237:19: note: ‘Mr_blocks’ was not initialized with a constant expression
  237 |     const int64_t Mr_blocks = (M + Mr - 1) / Mr;
      |                   ^~~~~~~~~
```
The PR also updates the UT to add a test for `BS`=512 in single thread.
The previous case has `BS`=1024 equal to the `K` and `N` value. The generated code does not have symbolic shapes thus fails to capture the above issue.
By adding a case of `BS`=512, the generated code will have symbolic shape for the M dim and is able to reproduce the issue that this PR is addressing.
Pull Request resolved: pytorch#136418
Approved by: https://github.com/jgong5

…no epilogue_nodes (pytorch#136518)
The `template_buffer_has_other_users` function checks the case where there're epilogue nodes and the template output has users other than these epilogue nodes.  When there's no epilogue nodes, the function could return `False` directly.
Pull Request resolved: pytorch#136518
Approved by: https://github.com/leslie-fang-intel, https://github.com/jgong5
ghstack dependencies: pytorch#136418

Fixes the max-autotune failure of `soft_actor_critic` of Torchbench in FP32 single thread dynamic shape case:
```log
  File "/home/user/inductor/pytorch/torch/_inductor/codegen/cpp_micro_gemm.py", line 136, in codegen_call
    C_ptr = f"&({kernel.index(C, [0, 0])})"
  File "/home/user/inductor/pytorch/torch/_inductor/codegen/cpp_template_kernel.py", line 135, in index
    else self.args.input(node.get_name())
  File "/home/user/inductor/pytorch/torch/_inductor/codegen/common.py", line 1251, in input
    assert name not in V.graph.removed_buffers, name
AssertionError: buf_GemmOut
```
The 1st and 2nd linear does not need to use local buffer while the 3rd linear needs to use local buffer.
The 3rd linear which uses local buffer will add its global buffer (named as `buf_GemmOut`) into `V.graph.removed_buffers`.
When scheduling the nodes, the 1st linear (won't use local buffer) will get its output buffer (also named as `buf_GemmOut`) from the input and found that it's in the `V.graph.removed_buffers` and raise AssertionError. The issue is that the output buffer of all these linears are all names with `buf_GemmOut`, which have a conflict.
Rename these buffers by adding the name of the `template_buffer` as the prefix.
Pull Request resolved: #136419
Approved by: https://github.com/leslie-fang-intel, https://github.com/jgong5
ghstack dependencies: #136418, #136518