mirrors/ZLUDA
0
0
Fork 0
mirror of https://github.com/vosen/ZLUDA.git synced 2025-08-03 14:50:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Add support for a new export table

Browse source
This commit is contained in:
Andrzej Janik 2021-09-12 15:12:26 +02:00
commit 5290190727
2 changed files with 89 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

79
zluda/src/impl/export_table.rs
View file

@ -2,6 +2,7 @@ use hip_runtime_sys::{
hipCtxCreate, hipDevicePrimaryCtxGetState, hipDevicePrimaryCtxRelease,
hipDevicePrimaryCtxRetain, hipError_t,
};
use ocl_core::ffi::c_uchar;
use crate::r#impl;
@ -51,6 +52,10 @@ pub fn get(table: *mut *const std::os::raw::c_void, id: *const CUuuid) -> CUresu
unsafe { *table = HEAP_ACCESS_VTABLE.as_ptr() as *const _ };
CUresult::CUDA_SUCCESS
}
DEVICE_EXTENDED_RT_GUID => {
unsafe { *table = DEVICE_EXTENDED_RT_VTABLE.as_ptr() as *const _ };
CUresult::CUDA_SUCCESS
}
_ => CUresult::CUDA_ERROR_NOT_SUPPORTED,
}
}
@ -504,3 +509,77 @@ unsafe extern "system" fn heap_alloc(
unsafe extern "system" fn heap_free(halloc: *mut HeapAllocRecord, arg1: *mut usize) -> CUresult {
r#impl::unimplemented()
}
const DEVICE_EXTENDED_RT_GUID: CUuuid = CUuuid {
bytes: [
0xB1u8, 0x05, 0x41, 0xE1, 0xF7, 0xC7, 0xC7, 0x4A, 0x9F, 0x64, 0xF2, 0x23, 0xBE, 0x99, 0xF1,
0xE2,
],
};
const DEVICE_EXTENDED_RT_LENGTH: usize = 21;
static DEVICE_EXTENDED_RT_VTABLE: [VTableEntry; DEVICE_EXTENDED_RT_LENGTH] = [
VTableEntry {
length: mem::size_of::<[VTableEntry; DEVICE_EXTENDED_RT_LENGTH]>(),
},
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry {
ptr: device_get_attribute_ext as _,
},
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: device_get_something as _ },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
VTableEntry { ptr: ptr::null() },
];
unsafe extern "system" fn device_get_attribute_ext(
_dev: CUdevice,
attribute: c_uint,
unknown: c_int,
result: *mut [usize; 2],
) -> CUresult {
if result == ptr::null_mut() {
return CUresult::CUDA_ERROR_INVALID_VALUE;
}
if unknown != 0 {
return CUresult::CUDA_ERROR_UNKNOWN;
}
// TODO: make real implementation
// Optix checks this probably toto know if HW RT is available
if attribute == 0x20000001 {
(&mut *result)[0] = 2;
(&mut *result)[1] = 0x130; // GTX 1080
} else if attribute == 0x20000002 {
(&mut *result)[0] = 2;
(&mut *result)[1] = 0x138; // GTX 1080
} else {
return CUresult::CUDA_ERROR_NOT_SUPPORTED;
}
CUresult::CUDA_SUCCESS
}
// I don't know is this function return,
// but on my GTX 1060 it returns 0
unsafe extern "system" fn device_get_something(
result: *mut c_uchar,
_dev: CUdevice,
) -> CUresult {
if result == ptr::null_mut() {
return CUresult::CUDA_ERROR_INVALID_VALUE;
}
*result = 0;
CUresult::CUDA_SUCCESS
}

11
zluda_dump/src/lib.rs
View file

@ -813,6 +813,13 @@ const HEAP_ACCESS_GUID: CUuuid = CUuuid {
],
};
const DEVICE_EXTENDED_RT_GUID: CUuuid = CUuuid {
bytes: [
0xB1u8, 0x05, 0x41, 0xE1, 0xF7, 0xC7, 0xC7, 0x4A, 0x9F, 0x64, 0xF2, 0x23, 0xBE, 0x99, 0xF1,
0xE2,
],
};
unsafe fn get_export_override_fn(
original_fn: *const c_void,
guid: *const CUuuid,
@ -828,7 +835,9 @@ unsafe fn get_export_override_fn(
| (CONTEXT_LOCAL_STORAGE_INTERFACE_V0301_GUID, 2)
| (CTX_CREATE_BYPASS_GUID, 1)
| (HEAP_ACCESS_GUID, 1)
| (HEAP_ACCESS_GUID, 2) => original_fn,
| (HEAP_ACCESS_GUID, 2)
| (DEVICE_EXTENDED_RT_GUID, 5)
| (DEVICE_EXTENDED_RT_GUID, 13) => original_fn,
(CUDART_INTERFACE_GUID, 1) => {
ORIGINAL_GET_MODULE_FROM_CUBIN = mem::transmute(original_fn);
get_module_from_cubin as *const _