Skip to content

Update for CUDA 13.0 #15

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Update for CUDA 13.0 #15

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6d9208c
Drop compute_70 arch
garcia Aug 9, 2025
2baa4b5
clockRate -> cudaDeviceGetAttribute(cudaDevAttrClockRate)
garcia Aug 9, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ endif()
set(FLAGS "")
set(CMAKE_CUDA_FLAGS "" CACHE STRING "" FORCE)
# Note that cmake will insert semicolons between these item automatically...
set(SM_LIST_IN "70,75,80,90" CACHE STRING "Please input compute capability")
set(SM_LIST_IN "75,80,90" CACHE STRING "Please input compute capability")
MARK_AS_ADVANCED(SM_LIST)
string(REPLACE "\," "\ " DISPLAY_SMS_LIST ${SM_LIST_IN})
string(REPLACE "\," "\;" SM_LIST ${SM_LIST_IN})
Expand Down
15 changes: 9 additions & 6 deletions util/measure.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -236,28 +236,31 @@ void printGemmSOL(int mathMode, double computeSeconds, int iterations, int m, in
return;
}

int clockRate;
cudaDeviceGetAttribute(&clockRate, cudaDevAttrClockRate, device_id);

// Set theoretical throughput to 0 at first (will be set later based on architecture)
double theoryThroughput = 0;

assert((prop.major == 3) || (prop.major == 5) || (prop.major == 6) || (prop.major == 7) || (prop.major == 8));
if(prop.major == 8) {
theoryThroughput = 2 * 64 * (double)prop.multiProcessorCount * (double)prop.clockRate*1e3;
theoryThroughput = 2 * 64 * (double)prop.multiProcessorCount * (double)clockRate*1e3;
} else if(prop.major == 7) {
theoryThroughput = 2 * 64 * (double)prop.multiProcessorCount * (double)prop.clockRate*1e3;
theoryThroughput = 2 * 64 * (double)prop.multiProcessorCount * (double)clockRate*1e3;
} else if(prop.major == 6) { // On Pascal, we have 64 or 128 FMAs per SM per clock
if(prop.minor == 0) { // SM60 GP100
theoryThroughput = 2 * 64 * (double)prop.multiProcessorCount * (double)prop.clockRate*1e3;
theoryThroughput = 2 * 64 * (double)prop.multiProcessorCount * (double)clockRate*1e3;
} else { // SM61+ GP102+
theoryThroughput = 2 * 128 * (double)prop.multiProcessorCount * (double)prop.clockRate*1e3;
theoryThroughput = 2 * 128 * (double)prop.multiProcessorCount * (double)clockRate*1e3;
}
}
// If Maxwell, we can compute 128FMAs per SM per clock
else if(prop.major > 3){
theoryThroughput = 2 * 128 * (double)prop.multiProcessorCount * (double)prop.clockRate*1e3;
theoryThroughput = 2 * 128 * (double)prop.multiProcessorCount * (double)clockRate*1e3;
}
// If Kepler, we can compute 192FMAs per SM per clock
else{
theoryThroughput = 2 * 192 * (double)prop.multiProcessorCount * (double)prop.clockRate*1e3;
theoryThroughput = 2 * 192 * (double)prop.multiProcessorCount * (double)clockRate*1e3;
}
// Correct for non-sgemm flops count, depending om the architecture
theoryThroughput *= coefGemmSOL<T_MATH>(mathMode, prop.major, prop.minor, algorithm);
Expand Down