gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fixes for building on Windows (#382)

Browse Source 
* make C++ code compatible with Windows + MSVC

* added dlfcn-win32 for cross-platform dlopen

* fixed building and pip install on Windows

* fixed shared library file name under Windows
This commit is contained in:
Victor
2021-12-07 14:10:58 -08:00
committed by GitHub
parent 0ff1a26b70
commit 73b04d71b2
10 changed files with 90 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
lib/codegen/selection/generator.cc
View File

@@ -441,18 +441,18 @@ std::tuple<Value*, Value*, Value*, Value*> generator::fp8x4_to_fp16x4(Value *in0
"lop3.b32 $1, b1, 0x80008000, a1, 0xf8; \n\t"
"}", "=r,=r,r", false);
Value *packed_in = UndefValue::get(vec_ty(i8_ty, 4));
packed_in = insert_elt(packed_in, in0, (int)0);
packed_in = insert_elt(packed_in, in1, (int)1);
packed_in = insert_elt(packed_in, in2, (int)2);
packed_in = insert_elt(packed_in, in3, (int)3);
packed_in = insert_elt(packed_in, in0, (uint64_t)0);
packed_in = insert_elt(packed_in, in1, (uint64_t)1);
packed_in = insert_elt(packed_in, in2, (uint64_t)2);
packed_in = insert_elt(packed_in, in3, (uint64_t)3);
Value *in = bit_cast(packed_in, i32_ty);
Value *ret = call(ptx, {in});
Value *packed_ret0 = extract_val(ret, {0});
Value *packed_ret1 = extract_val(ret, {1});
Value *ret0 = extract_elt(packed_ret0, (int)0);
Value *ret1 = extract_elt(packed_ret0, (int)1);
Value *ret2 = extract_elt(packed_ret1, (int)0);
Value *ret3 = extract_elt(packed_ret1, (int)1);
Value *ret0 = extract_elt(packed_ret0, (uint64_t)0);
Value *ret1 = extract_elt(packed_ret0, (uint64_t)1);
Value *ret2 = extract_elt(packed_ret1, (uint64_t)0);
Value *ret3 = extract_elt(packed_ret1, (uint64_t)1);
return std::make_tuple(ret0, ret1, ret2, ret3);
}
@@ -694,11 +694,11 @@ void generator::visit_load_inst(ir::load_inst* x){
// ---
// finally call inline ASM
// ---
InlineAsm *_asm = InlineAsm::get(asm_ty, asm_oss.str(), asm_cstrt, true);
InlineAsm *inlineAsm = InlineAsm::get(asm_ty, asm_oss.str(), asm_cstrt, true);
std::vector<Value*> args = {pred, ptr};
for(Value *v: others)
args.push_back(v);
Value *_ret = call(_asm, args);
Value *_ret = call(inlineAsm, args);
// ---
// extract and store return values
// ---

28
lib/driver/llvm.cc
View File

@@ -20,7 +20,9 @@
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#include <fstream>
#include <unistd.h>
#if __has_include(<unistd.h>)
#include <unistd.h>
#endif
#include <memory>
#include <regex>
#include "triton/driver/llvm.h"
@@ -165,10 +167,10 @@ std::string ptx_to_cubin(const std::string& ptx, int cc) {
return "";
// compile ptx with ptxas
char _fsrc[] = "/tmp/triton_k_XXXXXX";
char _flog[] = "/tmp/triton_l_XXXXXX";
mkstemp(_fsrc);
mkstemp(_flog);
char _fsrc[L_tmpnam];
char _flog[L_tmpnam];
std::tmpnam(_fsrc);
std::tmpnam(_flog);
std::string fsrc = _fsrc;
std::string flog = _flog;
std::string fbin = fsrc + ".o";
@@ -202,10 +204,10 @@ CUmodule ptx_to_cumodule(const std::string& ptx, int cc) {
// Use PTXAS via system call
if(use_system_ptxas){
// compile ptx with ptxas
char _fsrc[] = "/tmp/triton_k_XXXXXX";
char _flog[] = "/tmp/triton_l_XXXXXX";
mkstemp(_fsrc);
mkstemp(_flog);
char _fsrc[L_tmpnam];
char _flog[L_tmpnam];
std::tmpnam(_fsrc);
std::tmpnam(_flog);
std::string fsrc = _fsrc;
std::string flog = _flog;
std::string fbin = fsrc + ".o";
@@ -232,8 +234,8 @@ CUmodule ptx_to_cumodule(const std::string& ptx, int cc) {
CUjit_option opt[] = {CU_JIT_ERROR_LOG_BUFFER_SIZE_BYTES, CU_JIT_ERROR_LOG_BUFFER,
CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES, CU_JIT_INFO_LOG_BUFFER,
CU_JIT_LOG_VERBOSE};
unsigned int errbufsize = 8192;
unsigned int logbufsize = 8192;
const unsigned int errbufsize = 8192;
const unsigned int logbufsize = 8192;
char _err[errbufsize];
char _log[logbufsize];
void* optval[] = {(void*)(uintptr_t)errbufsize, (void*)_err, (void*)(uintptr_t)logbufsize, (void*)_log, (void*)1};
@@ -344,8 +346,8 @@ hipModule_t amdgpu_to_hipmodule(const std::string& path) {
hipJitOption opt[] = {hipJitOptionErrorLogBufferSizeBytes, hipJitOptionErrorLogBuffer,
hipJitOptionInfoLogBufferSizeBytes, hipJitOptionInfoLogBuffer,
hipJitOptionLogVerbose};
unsigned int errbufsize = 8192;
unsigned int logbufsize = 8192;
const unsigned int errbufsize = 8192;
const unsigned int logbufsize = 8192;
char _err[errbufsize];
char _log[logbufsize];
void* optval[] = {(void*)(uintptr_t)errbufsize,