FrozenGene commented on code in PR #14274:
URL: https://github.com/apache/tvm/pull/14274#discussion_r1144234538
##########
python/tvm/contrib/cc.py:
##########
@@ -231,6 +233,20 @@ def _linux_compile(output, objects, options, compile_cmd,
compile_shared=False):
else:
if compile_shared or output.endswith(".so") or
output.endswith(".dylib"):
cmd += ["--shared"]
+
+ compute_version = "".join(
+
get_target_compute_version(Target.current(allow_none=True)).split(".")
+ )
+ cmd += ["-gencode",
f"arch=compute_{compute_version},code=sm_{compute_version}"]
+ cmd += ["-O3"]
+ cmd += ["-std=c++17"]
+ cmd += ["-Xcompiler=-fPIC"]
+ cmd += ["-Xcompiler=-fno-strict-aliasing"]
+ cmd += ["-t 24"]
+ cmd += ["-DCUTLASS_ENABLE_TENSOR_CORE_MMA=1"]
+ cmd += ["-Xcompiler=-Wconversion"]
+ cmd += ["--use_fast_math"]
Review Comment:
doesn't make sense we enable fast math by default. We should control it
using one option and should set it false by default.
##########
python/tvm/contrib/cc.py:
##########
@@ -231,6 +233,20 @@ def _linux_compile(output, objects, options, compile_cmd,
compile_shared=False):
else:
if compile_shared or output.endswith(".so") or
output.endswith(".dylib"):
cmd += ["--shared"]
+
+ compute_version = "".join(
+
get_target_compute_version(Target.current(allow_none=True)).split(".")
+ )
+ cmd += ["-gencode",
f"arch=compute_{compute_version},code=sm_{compute_version}"]
+ cmd += ["-O3"]
+ cmd += ["-std=c++17"]
+ cmd += ["-Xcompiler=-fPIC"]
+ cmd += ["-Xcompiler=-fno-strict-aliasing"]
+ cmd += ["-t 24"]
Review Comment:
Should add the guard of CUDA version check >= 11.2 and to get the real cpu
cores: refer
this:https://github.com/apache/tvm/blob/main/python/tvm/contrib/cutlass/build.py#L72
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
