mirrors/ZLUDA
0
0
Fork 0
mirror of https://github.com/vosen/ZLUDA.git synced 2025-08-06 00:00:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Improve AMD compatibility

Browse source
This commit is contained in:
Andrzej Janik 2021-08-08 14:26:56 +02:00
commit e2432d0df1
2 changed files with 14 additions and 32 deletions
Download patch file Download diff file Expand all files Collapse all files

5
zluda/src/impl/function.rs
View file

@ -27,6 +27,7 @@ impl HasLivenessCookie for FunctionData {
pub struct FunctionData { pub struct FunctionData {
pub base: ocl_core::Kernel, pub base: ocl_core::Kernel,
pub device: ocl_core::DeviceId,
pub arg_size: Vec<(usize, bool)>, pub arg_size: Vec<(usize, bool)>,
pub use_shared_mem: bool, pub use_shared_mem: bool,
pub legacy_args: LegacyArguments, pub legacy_args: LegacyArguments,
@ -215,9 +216,9 @@ pub(crate) fn get_attribute(
CUfunction_attribute::CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK => { CUfunction_attribute::CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK => {
let max_threads = GlobalState::lock_function(func, |func| { let max_threads = GlobalState::lock_function(func, |func| {
if let ocl_core::KernelWorkGroupInfoResult::WorkGroupSize(size) = if let ocl_core::KernelWorkGroupInfoResult::WorkGroupSize(size) =
ocl_core::get_kernel_work_group_info::<()>( ocl_core::get_kernel_work_group_info(
&func.base, &func.base,
(), &func.device,
ocl_core::KernelWorkGroupInfo::WorkGroupSize, ocl_core::KernelWorkGroupInfo::WorkGroupSize,
)? )?
{ {

41
zluda/src/impl/module.rs
View file

@ -290,7 +290,7 @@ impl SpirvModule {
let binary = binary_prog.as_ptr(); let binary = binary_prog.as_ptr();
let mut binary_status = 0; let mut binary_status = 0;
let mut errcode_ret = 0; let mut errcode_ret = 0;
let program = unsafe { let raw_program = unsafe {
ocl_core::ffi::clCreateProgramWithBinary( ocl_core::ffi::clCreateProgramWithBinary(
ctx.as_ptr(), ctx.as_ptr(),
1, 1,
@ -303,7 +303,15 @@ impl SpirvModule {
}; };
assert_eq!(binary_status, 0, "clCreateProgramWithBinary"); assert_eq!(binary_status, 0, "clCreateProgramWithBinary");
assert_eq!(errcode_ret, 0, "clCreateProgramWithBinary"); assert_eq!(errcode_ret, 0, "clCreateProgramWithBinary");
unsafe { ocl_core::Program::from_raw_create_ptr(program) } let ocl_program = unsafe { ocl_core::Program::from_raw_create_ptr(raw_program) };
ocl_core::build_program(
&ocl_program,
Some(&[dev]),
&CString::new("").unwrap(),
None,
None,
)?;
ocl_program
} else { } else {
Self::compile_amd("gfx1011:xnack-", byte_il, self.should_link_ptx_impl).unwrap(); Self::compile_amd("gfx1011:xnack-", byte_il, self.should_link_ptx_impl).unwrap();
Self::compile_intel( Self::compile_intel(
@ -359,36 +367,9 @@ pub fn get_function(
&compiled_module.base, &compiled_module.base,
&entry.key().as_c_str().to_string_lossy(), &entry.key().as_c_str().to_string_lossy(),
)?; )?;
let true_b: ocl_core::ffi::cl_bool = 1;
let err = unsafe {
ocl_core::ffi::clSetKernelExecInfo(
kernel.as_ptr(),
CL_KERNEL_EXEC_INFO_INDIRECT_HOST_ACCESS_INTEL,
mem::size_of::<ocl_core::ffi::cl_bool>(),
&true_b as *const _ as *const _,
)
};
assert_eq!(err, 0);
let err = unsafe {
ocl_core::ffi::clSetKernelExecInfo(
kernel.as_ptr(),
CL_KERNEL_EXEC_INFO_INDIRECT_DEVICE_ACCESS_INTEL,
mem::size_of::<ocl_core::ffi::cl_bool>(),
&true_b as *const _ as *const _,
)
};
assert_eq!(err, 0);
let err = unsafe {
ocl_core::ffi::clSetKernelExecInfo(
kernel.as_ptr(),
CL_KERNEL_EXEC_INFO_INDIRECT_SHARED_ACCESS_INTEL,
mem::size_of::<ocl_core::ffi::cl_bool>(),
&true_b as *const _ as *const _,
)
};
assert_eq!(err, 0);
entry.insert(Box::new(Function::new(FunctionData { entry.insert(Box::new(Function::new(FunctionData {
base: kernel, base: kernel,
device: device.ocl_base.clone(),
arg_size: kernel_info.arguments_sizes.clone(), arg_size: kernel_info.arguments_sizes.clone(),
use_shared_mem: kernel_info.uses_shared_mem, use_shared_mem: kernel_info.uses_shared_mem,
legacy_args: LegacyArguments::new(), legacy_args: LegacyArguments::new(),