ggml-cuda: add mem check for fusion (#19916)
* ggml-cuda: add mem check for fusion * Replace NaNs with -FLT_MAX * fix typo Co-authored-by: Johannes Gäßler <johannesg@5d6.de> --------- Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
This commit is contained in:
co-authored by
Johannes Gäßler
parent
ba2ff79e43
commit
d48e876467
@@ -3412,6 +3412,69 @@ static bool ggml_cuda_can_fuse(const struct ggml_cgraph * cgraph,
|
|||||||
return false;
|
return false;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// returns whether the write (out) nodes overwrite the read nodes in operation
|
||||||
|
static bool ggml_cuda_check_fusion_memory_ranges(ggml_cgraph * cgraph,
|
||||||
|
int node_idx,
|
||||||
|
int node_count,
|
||||||
|
int * out_nodes,
|
||||||
|
int out_count) {
|
||||||
|
auto nodes_overlap = [&](const ggml_tensor * a, const ggml_tensor * b) {
|
||||||
|
const int64_t a_start = (int64_t) a->data;
|
||||||
|
const int64_t a_end = a_start + ggml_nbytes(a);
|
||||||
|
|
||||||
|
const int64_t b_start = (int64_t) b->data;
|
||||||
|
const int64_t b_end = b_start + ggml_nbytes(b);
|
||||||
|
|
||||||
|
if ((b_start <= a_start && a_start < b_end) || (a_start <= b_start && b_start < a_end)) {
|
||||||
|
return true;
|
||||||
|
}
|
||||||
|
|
||||||
|
return false;
|
||||||
|
};
|
||||||
|
|
||||||
|
bool is_ok = true;
|
||||||
|
// for nrows=1, all fusion operations correctly read the src before writing dst or do it elementwise, so we should be ok
|
||||||
|
if (ggml_nrows(cgraph->nodes[node_idx]) == 1) {
|
||||||
|
return true;
|
||||||
|
}
|
||||||
|
|
||||||
|
for (int i = 0; i < out_count; ++i) {
|
||||||
|
const ggml_tensor * dst = cgraph->nodes[out_nodes[i]];
|
||||||
|
|
||||||
|
for (int j = node_idx; j < node_idx + node_count; ++j) {
|
||||||
|
// Loop over all srcs of all nodes in the fusion. If the src overlaps
|
||||||
|
// the destination and the src is not an intermediate node that's being
|
||||||
|
// elided, then disable fusion.
|
||||||
|
|
||||||
|
for (int src_idx = 0; src_idx < GGML_MAX_SRC; ++src_idx) {
|
||||||
|
const ggml_tensor * src = cgraph->nodes[j]->src[src_idx];
|
||||||
|
|
||||||
|
if (!src || src->op == GGML_OP_NONE) {
|
||||||
|
continue;
|
||||||
|
}
|
||||||
|
|
||||||
|
if (nodes_overlap(dst, src)) {
|
||||||
|
bool found = false;
|
||||||
|
|
||||||
|
for (int k = node_idx; k < j; ++k) {
|
||||||
|
if (cgraph->nodes[k] == src) {
|
||||||
|
found = true;
|
||||||
|
break;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
if (!found) {
|
||||||
|
is_ok = false;
|
||||||
|
break;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
return is_ok;
|
||||||
|
}
|
||||||
|
|
||||||
static void ggml_cuda_graph_evaluate_and_capture(ggml_backend_cuda_context * cuda_ctx, ggml_cgraph * cgraph, const bool use_cuda_graph, const bool cuda_graph_update_required, const void * graph_key) {
|
static void ggml_cuda_graph_evaluate_and_capture(ggml_backend_cuda_context * cuda_ctx, ggml_cgraph * cgraph, const bool use_cuda_graph, const bool cuda_graph_update_required, const void * graph_key) {
|
||||||
bool graph_evaluated_or_captured = false;
|
bool graph_evaluated_or_captured = false;
|
||||||
|
|
||||||
@@ -3608,7 +3671,8 @@ static void ggml_cuda_graph_evaluate_and_capture(ggml_backend_cuda_context * cud
|
|||||||
out_nodes[1] = i + ops.size() - 1;
|
out_nodes[1] = i + ops.size() - 1;
|
||||||
|
|
||||||
if (ggml_can_fuse_subgraph(cgraph, i, ops.size(), ops.data(), out_nodes, 2) &&
|
if (ggml_can_fuse_subgraph(cgraph, i, ops.size(), ops.data(), out_nodes, 2) &&
|
||||||
ggml_cuda_should_use_topk_moe(node, logits, weights, ids)) {
|
ggml_cuda_should_use_topk_moe(node, logits, weights, ids) &&
|
||||||
|
ggml_cuda_check_fusion_memory_ranges(cgraph, i, ops.size(), out_nodes, 2)) {
|
||||||
ggml_cuda_op_topk_moe(*cuda_ctx, logits, weights, ids, clamp, scale, bias, args);
|
ggml_cuda_op_topk_moe(*cuda_ctx, logits, weights, ids, clamp, scale, bias, args);
|
||||||
i += ops.size() - 1;
|
i += ops.size() - 1;
|
||||||
continue;
|
continue;
|
||||||
@@ -3623,7 +3687,8 @@ static void ggml_cuda_graph_evaluate_and_capture(ggml_backend_cuda_context * cud
|
|||||||
|
|
||||||
int out_nodes[2] = { i + 1, i + 5 };
|
int out_nodes[2] = { i + 1, i + 5 };
|
||||||
if (ggml_can_fuse_subgraph(cgraph, i, ops.size(), ops.data(), out_nodes, 2) &&
|
if (ggml_can_fuse_subgraph(cgraph, i, ops.size(), ops.data(), out_nodes, 2) &&
|
||||||
ggml_cuda_should_use_topk_moe(softmax, logits, weights, ids)) {
|
ggml_cuda_should_use_topk_moe(softmax, logits, weights, ids) &&
|
||||||
|
ggml_cuda_check_fusion_memory_ranges(cgraph, i, ops.size(), out_nodes, 2)) {
|
||||||
ggml_cuda_op_topk_moe(*cuda_ctx, logits, weights, ids, clamp, scale, bias, args);
|
ggml_cuda_op_topk_moe(*cuda_ctx, logits, weights, ids, clamp, scale, bias, args);
|
||||||
i += ops.size() - 1;
|
i += ops.size() - 1;
|
||||||
continue;
|
continue;
|
||||||
|
|||||||
@@ -119,6 +119,18 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// Sanitize NaN to -FLT_MAX so the iterative argmax produces unique expert IDs.
|
||||||
|
// NaN comparisons always return false, which would cause the same expert to be
|
||||||
|
// selected repeatedly. -FLT_MAX compares normally and is still excluded by the
|
||||||
|
// -INFINITY sentinel used after each selection round.
|
||||||
|
// More relevant for the cuBLAS path. See https://github.com/ggml-org/llama.cpp/issues/19659
|
||||||
|
#pragma unroll
|
||||||
|
for (int i = 0; i < experts_per_thread; i++) {
|
||||||
|
if (__isnanf(wt[i])) {
|
||||||
|
wt[i] = -FLT_MAX;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
// selection_wt is only needed when bias is present (selection uses wt + bias)
|
// selection_wt is only needed when bias is present (selection uses wt + bias)
|
||||||
// when no bias, we use wt directly for both selection and weight values
|
// when no bias, we use wt directly for both selection and weight values
|
||||||
float selection_wt[has_bias ? experts_per_thread : 1];
|
float selection_wt[has_bias ? experts_per_thread : 1];
|
||||||
|
|||||||
Reference in New Issue
Block a user