DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-12 11:56:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor: reuse ggml internals in backend extensions

Browse Source
This commit is contained in:
leejet committed 2026-11-01 04:01:26 +08:00
1 parent 95773aa216
commit 6cfc3572f7
26 files changed
+358 -590
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/core/cuda/affine.cu
+6 -6
View File
@@ -1,8 +1,8 @@
#include <climits>
#include <cstdlib>
#include <cstring>
#include "backend.h"
#include "ggml-ext.h"
#include "runtime.h"
namespace SD_CUDA_NAMESPACE {
namespace sd_affine_cuda {
@@ -129,7 +129,7 @@ namespace SD_CUDA_NAMESPACE {
ggml_ext_gpu_context context;
if (!cuda_context(execution->backend, &context))
return GGML_STATUS_FAILED;
auto stream = static_cast<CUDAStream>(context.stream);
auto stream = static_cast<cudaStream_t>(context.stream);
const auto* node = nodes[0];
const auto* a = node->src[0];
const auto* b = node->op == GGML_OP_CONT ? nullptr : node->src[1];
@@ -154,10 +154,10 @@ namespace SD_CUDA_NAMESPACE {
scale_permute<<<blocks, 256, 0, stream>>>(static_cast<const float*>(a->data), static_cast<half2*>(output),
a->ne[0], a->ne[2], a->ne[1], ggml_nelements(dst), a->nb[1] / 4, a->nb[2] / 4, a->nb[3] / 4, param(nodes[1], 0));
}
auto error = cuda_last_error();
if (error == cuda_success && alias)
error = cuda_copy_async(dst->data, output, ggml_nbytes(dst), stream);
return error == cuda_success ? GGML_STATUS_SUCCESS : GGML_STATUS_FAILED;
auto error = cudaGetLastError();
if (error == cudaSuccess && alias)
error = cudaMemcpyAsync(dst->data, output, ggml_nbytes(dst), cudaMemcpyDeviceToDevice, stream);
return error == cudaSuccess ? GGML_STATUS_SUCCESS : GGML_STATUS_FAILED;
}
}