Fix host code and update to CUDA 12.4 (#299)

1 files changed, 30 insertions, 50 deletions

diff --git a/zluda/src/impl/memory.rs b/zluda/src/impl/memory.rs
index 6041623..3843776 100644
--- a/zluda/src/impl/memory.rs
+++ b/zluda/src/impl/memory.rs
@@ -1,55 +1,35 @@
-use hip_runtime_sys::{
-    hipDrvMemcpy3D, hipError_t, hipMemcpy3D, hipMemcpy3DParms, hipMemoryType, hipPitchedPtr,
-    hipPos, HIP_MEMCPY3D,
-};
-use std::ptr;
+use hip_runtime_sys::*;
 
-use crate::{
-    cuda::{CUDA_MEMCPY3D_st, CUdeviceptr, CUmemorytype, CUresult},
-    hip_call,
-};
+pub(crate) fn alloc_v2(dptr: *mut hipDeviceptr_t, bytesize: usize) -> hipError_t {
+    unsafe { hipMalloc(dptr.cast(), bytesize) }?;
+    // TODO: parametrize for non-Geekbench
+    unsafe { hipMemsetD8(*dptr, 0, bytesize) }
+}
+
+pub(crate) fn free_v2(dptr: hipDeviceptr_t) -> hipError_t {
+    unsafe { hipFree(dptr.0) }
+}
+
+pub(crate) fn copy_dto_h_v2(
+    dst_host: *mut ::core::ffi::c_void,
+    src_device: hipDeviceptr_t,
+    byte_count: usize,
+) -> hipError_t {
+    unsafe { hipMemcpyDtoH(dst_host, src_device, byte_count) }
+}
 
-// TODO change HIP impl to 64 bits
-pub(crate) unsafe fn copy_3d(cu_copy: *const CUDA_MEMCPY3D_st) -> Result<(), hipError_t> {
-    if cu_copy == ptr::null() {
-        return Err(hipError_t::hipErrorInvalidValue);
-    }
-    let cu_copy = *cu_copy;
-    let hip_copy = HIP_MEMCPY3D {
-        srcXInBytes: cu_copy.srcXInBytes as u32,
-        srcY: cu_copy.srcY as u32,
-        srcZ: cu_copy.srcZ as u32,
-        srcLOD: cu_copy.srcLOD as u32,
-        srcMemoryType: memory_type(cu_copy.srcMemoryType)?,
-        srcHost: cu_copy.srcHost,
-        srcDevice: cu_copy.srcDevice.0 as _,
-        srcArray: cu_copy.srcArray as _,
-        srcPitch: cu_copy.srcPitch as u32,
-        srcHeight: cu_copy.srcHeight as u32,
-        dstXInBytes: cu_copy.dstXInBytes as u32,
-        dstY: cu_copy.dstY as u32,
-        dstZ: cu_copy.dstZ as u32,
-        dstLOD: cu_copy.dstLOD as u32,
-        dstMemoryType: memory_type(cu_copy.dstMemoryType)?,
-        dstHost: cu_copy.dstHost,
-        dstDevice: cu_copy.dstDevice.0 as _,
-        dstArray: cu_copy.dstArray as _,
-        dstPitch: cu_copy.dstPitch as u32,
-        dstHeight: cu_copy.dstHeight as u32,
-        WidthInBytes: cu_copy.WidthInBytes as u32,
-        Height: cu_copy.Height as u32,
-        Depth: cu_copy.Depth as u32,
-    };
-    hip_call! { hipDrvMemcpy3D(&hip_copy) };
-    Ok(())
+pub(crate) fn copy_hto_d_v2(
+    dst_device: hipDeviceptr_t,
+    src_host: *const ::core::ffi::c_void,
+    byte_count: usize,
+) -> hipError_t {
+    unsafe { hipMemcpyHtoD(dst_device, src_host.cast_mut(), byte_count) }
 }
 
-pub(crate) fn memory_type(cu: CUmemorytype) -> Result<hipMemoryType, hipError_t> {
-    match cu {
-        CUmemorytype::CU_MEMORYTYPE_HOST => Ok(hipMemoryType::hipMemoryTypeHost),
-        CUmemorytype::CU_MEMORYTYPE_DEVICE => Ok(hipMemoryType::hipMemoryTypeDevice),
-        CUmemorytype::CU_MEMORYTYPE_ARRAY => Ok(hipMemoryType::hipMemoryTypeArray),
-        CUmemorytype::CU_MEMORYTYPE_UNIFIED => Ok(hipMemoryType::hipMemoryTypeUnified),
-        _ => Err(hipError_t::hipErrorInvalidValue),
-    }
+pub(crate) fn get_address_range_v2(
+    pbase: *mut hipDeviceptr_t,
+    psize: *mut usize,
+    dptr: hipDeviceptr_t,
+) -> hipError_t {
+    unsafe { hipMemGetAddressRange(pbase, psize, dptr) }
 }