Optimize the kernel module to inline the actual kernel into wrapper.

Diptorup Deb · Diptorup Deb · commit d3d7ef67cff4 · 2023-08-17T21:01:52.000-05:00
diff --git a/numba_dpex/core/codegen.py b/numba_dpex/core/codegen.py
@@ -35,6 +35,7 @@ def _optimize_final_module(self):
         pmb.opt_level = config.OPT
 
         pmb.disable_unit_at_a_time = False
+        pmb.inlining_threshold = 2
         pmb.disable_unroll_loops = True
         pmb.loop_vectorize = False
         pmb.slp_vectorize = False
diff --git a/numba_dpex/core/kernel_interface/spirv_kernel.py b/numba_dpex/core/kernel_interface/spirv_kernel.py
@@ -136,6 +136,7 @@ def compile(
         kernel = cres.target_context.prepare_ocl_kernel(
             func, cres.signature.args
         )
+        cres.library._optimize_final_module()
         self._llvm_module = kernel.module.__str__()
         self._module_name = kernel.name
 

Original file line number	Diff line number	Diff line change
`@@ -136,6 +136,7 @@ def compile(`
`136`	`136`	`kernel = cres.target_context.prepare_ocl_kernel(`
`137`	`137`	`func, cres.signature.args`
`138`	`138`	`)`
	`139`	`+ cres.library._optimize_final_module()`
`139`	`140`	`self._llvm_module = kernel.module.__str__()`
`140`	`141`	`self._module_name = kernel.name`
`141`	`142`