Skip to content

ggml-cuda: avoid direct ROCm_Host compute on HIP integrated GPUs (port of ggml-org#25863) #158

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
danielhanchen wants to merge 1 commit into
base: master
Choose a base branch
from
+25 −2
Open

ggml-cuda: avoid direct ROCm_Host compute on HIP integrated GPUs (port of ggml-org#25863) #158

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 25 additions & 2 deletions ggml/src/ggml-cuda/ggml-cuda.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4822,6 +4822,22 @@ static enum ggml_backend_dev_type ggml_backend_cuda_device_get_type(ggml_backend
: GGML_BACKEND_DEVICE_TYPE_GPU;
}

static bool ggml_backend_cuda_host_buffer_supported() {
return getenv("GGML_CUDA_NO_PINNED") == nullptr;
}

static bool ggml_backend_cuda_device_supports_cuda_host_buft(int device) {
#if defined(GGML_USE_HIP)
if (ggml_cuda_info().devices[device].integrated) {
return false;
}
#else
GGML_UNUSED(device);
#endif

return ggml_backend_cuda_host_buffer_supported();
}

static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_backend_dev_props * props) {
ggml_backend_cuda_device_context * ctx = (ggml_backend_cuda_device_context *)dev->context;

Expand All @@ -4831,7 +4847,7 @@ static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_back
props->device_id = ctx->pci_bus_id.empty() ? nullptr : ctx->pci_bus_id.c_str();
ggml_backend_cuda_device_get_memory(dev, &props->memory_free, &props->memory_total);

bool host_buffer = getenv("GGML_CUDA_NO_PINNED") == nullptr;
bool host_buffer = ggml_backend_cuda_host_buffer_supported();
#ifdef GGML_CUDA_NO_PEER_COPY
bool events = false;
#else
Expand Down Expand Up @@ -4860,6 +4876,10 @@ static ggml_backend_buffer_type_t ggml_backend_cuda_device_get_buffer_type(ggml_

static ggml_backend_buffer_type_t ggml_backend_cuda_device_get_host_buffer_type(ggml_backend_dev_t dev) {
GGML_UNUSED(dev);
if (!ggml_backend_cuda_host_buffer_supported()) {
return nullptr;
}

return ggml_backend_cuda_host_buffer_type();
}

Expand Down Expand Up @@ -5320,7 +5340,10 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
static bool ggml_backend_cuda_device_supports_buft(ggml_backend_dev_t dev, ggml_backend_buffer_type_t buft) {
ggml_backend_cuda_device_context * dev_ctx = (ggml_backend_cuda_device_context *) dev->context;
const bool integrated = ggml_cuda_info().devices[dev_ctx->device].integrated;
return (ggml_backend_buft_is_cuda(buft) && buft->device == dev) || (integrated && ggml_backend_buft_is_cuda_host(buft));
return (ggml_backend_buft_is_cuda(buft) && buft->device == dev) ||
(integrated &&
ggml_backend_buft_is_cuda_host(buft) &&
ggml_backend_cuda_device_supports_cuda_host_buft(dev_ctx->device));
}

static int64_t get_op_batch_size(const ggml_tensor * op) {
Expand Down