JinjieLiu/triton-tvm-ffi
1
0
Fork 0
mirror of https://github.com/sgjzfzzf/triton-tvm-ffi.git synced 2026-05-02 03:52:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

support mm and autotune

Browse Source 
Signed-off-by: jinjieliu <jinjie.liu@usc.edu>
This commit is contained in:
jinjieliu
2026-02-07 00:41:23 +08:00
parent f6c7a48c1b
commit 2298b6f8c8
7 changed files with 486 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
examples/add/add.cc
View File

@@ -5,8 +5,7 @@
#include <tvm/ffi/tvm_ffi.h>
#ifndef ADD_KERNEL_STUB
#define ADD_KERNEL_STUB(grid, stream, numWarps, numStages, x, y, output, \
numel, BLOCK_SIZE)
#define ADD_KERNEL_STUB(grid, stream, numWarps, numStages, args, kwargs)
#endif
#ifndef ADD_NAME
@@ -27,7 +26,9 @@ tvm::ffi::Tensor Add(tvm::ffi::Tensor x, tvm::ffi::Tensor y) {
tvm::ffi::Optional<int32_t> numWarps = std::nullopt, numStages = std::nullopt;
DLDevice device = x.device();
void *stream = TVMFFIEnvGetStream(device.device_type, device.device_id);
ADD_KERNEL_STUB(grid, stream, numWarps, numStages, x, y, output, numel, 1024);
tvm::ffi::Array<tvm::ffi::Any> args = {x, y, output, numel, 1024};
tvm::ffi::Map<tvm::ffi::String, tvm::ffi::Any> kwargs = {};
ADD_KERNEL_STUB(grid, stream, numWarps, numStages, args, kwargs);
return output;
}