Gather qmm batched kernel and refactoring of quantized (#2078)

This commit is contained in:
Angelos Katharopoulos
2025-04-17 13:53:11 -07:00
committed by GitHub
parent 99eefd2ec0
commit 5de6d94a90
15 changed files with 1479 additions and 449 deletions

View File

@@ -269,4 +269,21 @@ MTL::ComputePipelineState* get_quantized_kernel(
return d.get_kernel(kernel_name);
}
MTL::ComputePipelineState* get_gather_qmm_kernel(
metal::Device& d,
const std::string& kernel_name,
const std::string& hash_name,
const metal::MTLFCList& func_consts,
const array&,
int,
int,
int,
int,
int,
int,
int,
bool) {
return d.get_kernel(kernel_name, "mlx", hash_name, func_consts);
}
} // namespace mlx::core