Use CU_MEMCPY_SRC_ACCESS_ORDER_ANY for batch KV cache swaps (#39306)
Signed-off-by: Itay Etelis <itay.etelis@ibm.com> Signed-off-by: Itay Etelis <etelis2019@gmail.com> Signed-off-by: Itay Etelis <92247226+Etelis@users.noreply.github.com> Co-authored-by: Itay Etelis <itay.etelis@ibm.com> Co-authored-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Itay Etelis <etelis2019@gmail.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
co-authored by
Itay Etelis
Or Ozeri
Itay Etelis
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent
0d382ecde8
commit
00b0618a03
+2
-1
@@ -12,7 +12,8 @@ void swap_blocks(torch::Tensor& src, torch::Tensor& dst,
|
|||||||
|
|
||||||
void swap_blocks_batch(const torch::Tensor& src_ptrs,
|
void swap_blocks_batch(const torch::Tensor& src_ptrs,
|
||||||
const torch::Tensor& dst_ptrs,
|
const torch::Tensor& dst_ptrs,
|
||||||
const torch::Tensor& sizes);
|
const torch::Tensor& sizes,
|
||||||
|
bool is_src_access_order_any);
|
||||||
|
|
||||||
void reshape_and_cache(torch::Tensor& key, torch::Tensor& value,
|
void reshape_and_cache(torch::Tensor& key, torch::Tensor& value,
|
||||||
torch::Tensor& key_cache, torch::Tensor& value_cache,
|
torch::Tensor& key_cache, torch::Tensor& value_cache,
|
||||||
|
|||||||
@@ -77,7 +77,8 @@ void swap_blocks(torch::Tensor& src, torch::Tensor& dst,
|
|||||||
|
|
||||||
void swap_blocks_batch(const torch::Tensor& src_ptrs,
|
void swap_blocks_batch(const torch::Tensor& src_ptrs,
|
||||||
const torch::Tensor& dst_ptrs,
|
const torch::Tensor& dst_ptrs,
|
||||||
const torch::Tensor& sizes) {
|
const torch::Tensor& sizes,
|
||||||
|
bool is_src_access_order_any) {
|
||||||
TORCH_CHECK(src_ptrs.device().is_cpu(), "src_ptrs must be on CPU");
|
TORCH_CHECK(src_ptrs.device().is_cpu(), "src_ptrs must be on CPU");
|
||||||
TORCH_CHECK(dst_ptrs.device().is_cpu(), "dst_ptrs must be on CPU");
|
TORCH_CHECK(dst_ptrs.device().is_cpu(), "dst_ptrs must be on CPU");
|
||||||
TORCH_CHECK(sizes.device().is_cpu(), "sizes must be on CPU");
|
TORCH_CHECK(sizes.device().is_cpu(), "sizes must be on CPU");
|
||||||
@@ -124,7 +125,12 @@ void swap_blocks_batch(const torch::Tensor& src_ptrs,
|
|||||||
|
|
||||||
if (batch_fn != nullptr) {
|
if (batch_fn != nullptr) {
|
||||||
CUmemcpyAttributes attr = {};
|
CUmemcpyAttributes attr = {};
|
||||||
attr.srcAccessOrder = CU_MEMCPY_SRC_ACCESS_ORDER_STREAM;
|
// ANY lets the DMA engine prefetch source bytes out of stream order,
|
||||||
|
// which is only safe when no GPU stream is concurrently writing the
|
||||||
|
// source.
|
||||||
|
attr.srcAccessOrder = is_src_access_order_any
|
||||||
|
? CU_MEMCPY_SRC_ACCESS_ORDER_ANY
|
||||||
|
: CU_MEMCPY_SRC_ACCESS_ORDER_STREAM;
|
||||||
size_t attrs_idx = 0;
|
size_t attrs_idx = 0;
|
||||||
size_t fail_idx = 0;
|
size_t fail_idx = 0;
|
||||||
CUresult result = batch_fn(reinterpret_cast<CUdeviceptr*>(dst_data),
|
CUresult result = batch_fn(reinterpret_cast<CUdeviceptr*>(dst_data),
|
||||||
|
|||||||
@@ -553,7 +553,8 @@ TORCH_LIBRARY_EXPAND(CONCAT(TORCH_EXTENSION_NAME, _cache_ops), cache_ops) {
|
|||||||
// Batch swap: submit all block copies in a single driver call.
|
// Batch swap: submit all block copies in a single driver call.
|
||||||
cache_ops.def(
|
cache_ops.def(
|
||||||
"swap_blocks_batch(Tensor src_ptrs, Tensor dst_ptrs,"
|
"swap_blocks_batch(Tensor src_ptrs, Tensor dst_ptrs,"
|
||||||
" Tensor sizes) -> ()");
|
" Tensor sizes,"
|
||||||
|
" bool is_src_access_order_any=False) -> ()");
|
||||||
cache_ops.impl("swap_blocks_batch", torch::kCPU, &swap_blocks_batch);
|
cache_ops.impl("swap_blocks_batch", torch::kCPU, &swap_blocks_batch);
|
||||||
|
|
||||||
// Reshape the key and value tensors and cache them.
|
// Reshape the key and value tensors and cache them.
|
||||||
|
|||||||
+10
-1
@@ -2805,6 +2805,7 @@ def swap_blocks_batch(
|
|||||||
src_ptrs: torch.Tensor,
|
src_ptrs: torch.Tensor,
|
||||||
dst_ptrs: torch.Tensor,
|
dst_ptrs: torch.Tensor,
|
||||||
sizes: torch.Tensor,
|
sizes: torch.Tensor,
|
||||||
|
is_src_access_order_any: bool = False,
|
||||||
) -> None:
|
) -> None:
|
||||||
"""
|
"""
|
||||||
Batch version of swap_blocks: submit all copies in a single driver call.
|
Batch version of swap_blocks: submit all copies in a single driver call.
|
||||||
@@ -2813,8 +2814,16 @@ def swap_blocks_batch(
|
|||||||
of sizes[i] bytes. All three tensors must be int64 CPU tensors.
|
of sizes[i] bytes. All three tensors must be int64 CPU tensors.
|
||||||
On CUDA 12.8+ this uses cuMemcpyBatchAsync for minimal submission
|
On CUDA 12.8+ this uses cuMemcpyBatchAsync for minimal submission
|
||||||
overhead; on older CUDA it falls back to a loop of cudaMemcpyAsync.
|
overhead; on older CUDA it falls back to a loop of cudaMemcpyAsync.
|
||||||
|
|
||||||
|
is_src_access_order_any: if True, pass CU_MEMCPY_SRC_ACCESS_ORDER_ANY to
|
||||||
|
cuMemcpyBatchAsync, letting the DMA engine prefetch source bytes
|
||||||
|
out of stream order. Only safe when no GPU stream is concurrently
|
||||||
|
writing to the source. Defaults to False (STREAM ordering), which
|
||||||
|
is always safe.
|
||||||
"""
|
"""
|
||||||
torch.ops._C_cache_ops.swap_blocks_batch(src_ptrs, dst_ptrs, sizes)
|
torch.ops._C_cache_ops.swap_blocks_batch(
|
||||||
|
src_ptrs, dst_ptrs, sizes, is_src_access_order_any
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
def convert_fp8(
|
def convert_fp8(
|
||||||
|
|||||||
@@ -313,10 +313,22 @@ class SingleDirectionOffloadingHandler(OffloadingHandler):
|
|||||||
last_event = last_transfer.end_event
|
last_event = last_transfer.end_event
|
||||||
# assure job will start only after the previous one completes
|
# assure job will start only after the previous one completes
|
||||||
stream.wait_event(last_event)
|
stream.wait_event(last_event)
|
||||||
|
# CPU->GPU reads from host pinned memory, which is never written
|
||||||
|
# by a concurrent GPU stream, so CU_MEMCPY_SRC_ACCESS_ORDER_ANY is
|
||||||
|
# safe and lets the driver pipeline source reads. GPU->CPU reads
|
||||||
|
# from the live GPU KV cache, which the compute stream keeps
|
||||||
|
# writing; we must keep STREAM ordering so source reads are gated
|
||||||
|
# by the transfer stream's wait_stream(compute) barrier.
|
||||||
|
is_src_access_order_any = not self.gpu_to_cpu
|
||||||
with torch.cuda.stream(stream):
|
with torch.cuda.stream(stream):
|
||||||
start_event.record(stream)
|
start_event.record(stream)
|
||||||
if num_copy_ops > 0:
|
if num_copy_ops > 0:
|
||||||
ops.swap_blocks_batch(batch_src, batch_dst, batch_sizes)
|
ops.swap_blocks_batch(
|
||||||
|
batch_src,
|
||||||
|
batch_dst,
|
||||||
|
batch_sizes,
|
||||||
|
is_src_access_order_any=is_src_access_order_any,
|
||||||
|
)
|
||||||
end_event.record(stream)
|
end_event.record(stream)
|
||||||
|
|
||||||
self._transfer_events[job_id] = end_event
|
self._transfer_events[job_id] = end_event
|
||||||
|
|||||||
Reference in New Issue
Block a user