Skip to content

Commit a3a5230

Browse files
ikawrakowIwan Kawrakow
ikawrakow
and
Iwan Kawrakow
authored
Revert "Better CPU prompt processing performance for SWA models (#696)" (#701)
This reverts commit 93a4f60. Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
1 parent 7d14f8e commit a3a5230

File tree

5 files changed

+30
-140
lines changed

5 files changed

+30
-140
lines changed

‎ggml/include/ggml.h‎

Lines changed: 0 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -2043,10 +2043,6 @@ extern "C" {
20432043
struct ggml_tensor * a,
20442044
struct ggml_tensor * sinks);
20452045

2046-
GGML_API void ggml_flash_attn_ext_add_bounds(
2047-
struct ggml_tensor * a,
2048-
struct ggml_tensor * bounds);
2049-
20502046
// TODO: needs to be adapted to ggml_flash_attn_ext
20512047
GGML_API struct ggml_tensor * ggml_flash_attn_back(
20522048
struct ggml_context * ctx,

‎ggml/src/ggml.c‎

Lines changed: 1 addition & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -8993,22 +8993,6 @@ void ggml_flash_attn_ext_add_sinks(
89938993
a->src[4] = sinks;
89948994
}
89958995

8996-
void ggml_flash_attn_ext_add_bounds(
8997-
struct ggml_tensor * a,
8998-
struct ggml_tensor * bounds) {
8999-
if (!bounds) {
9000-
a->src[5] = NULL;
9001-
return;
9002-
}
9003-
9004-
GGML_ASSERT(a->op == GGML_OP_FLASH_ATTN_EXT);
9005-
GGML_ASSERT(bounds->type == GGML_TYPE_I32);
9006-
GGML_ASSERT(bounds->ne[0] == 2);
9007-
GGML_ASSERT(bounds->ne[1] >= a->src[0]->ne[1]);
9008-
9009-
a->src[5] = bounds;
9010-
}
9011-
90128996
// ggml_flash_attn_back
90138997

90148998
struct ggml_tensor * ggml_flash_attn_back(
@@ -18677,7 +18661,6 @@ static void ggml_compute_forward_flash_attn_ext_f16(
1867718661
const struct ggml_tensor * v = dst->src[2];
1867818662
const struct ggml_tensor * mask = dst->src[3];
1867918663
const struct ggml_tensor * sinks = dst->src[4];
18680-
const struct ggml_tensor * bounds= dst->src[5];
1868118664

1868218665
GGML_TENSOR_LOCALS(int64_t, neq, q, ne)
1868318666
GGML_TENSOR_LOCALS(size_t, nbq, q, nb)
@@ -18756,9 +18739,7 @@ static void ggml_compute_forward_flash_attn_ext_f16(
1875618739
dst->ne[2], dst->ne[1], dst->nb[1],
1875718740
k->type, v->type,
1875818741
Dk, Dv, neq1, nek1, q->nb[1], k->nb[1], v->nb[1], mask->nb[1],
18759-
q->data, k->data, v->data, mask->data,
18760-
sinks ? sinks->data : NULL,
18761-
bounds ? bounds->data : NULL,
18742+
q->data, k->data, v->data, mask->data, sinks ? sinks->data : NULL,
1876218743
scale, softcap, (float *)dst->data,
1876318744
params->wdata, (barrier_t)ggml_barrier, (void *)params->shared, ith, nth)) return;
1876418745

‎ggml/src/iqk/iqk_flash_attn.cpp‎

Lines changed: 18 additions & 50 deletions
Original file line numberDiff line numberDiff line change
@@ -43,27 +43,6 @@ inline void accumulate_qkv(int Dv, float& M, float& S, float Mj, float Sj, float
4343
for (int i = 0; i < Dv; ++i) Racc[i] += c*R[i];
4444
}
4545
}
46-
inline std::pair<int, int> mask_range(int nek1, const uint16_t * umask) {
47-
int first_k = 0, last_k = nek1;
48-
for (; first_k < last_k; ++first_k) {
49-
if (umask[first_k] == 0) break;
50-
}
51-
for (; last_k > first_k; --last_k) {
52-
if (umask[last_k-1] == 0) break;
53-
}
54-
return { first_k, last_k };
55-
}
56-
inline bool reduce_k_range(int nek1, int& first_k, int& last_k) {
57-
int nk = last_k - first_k;
58-
if (nk >= nek1) return false;
59-
if (nk%32) {
60-
int nk32 = 32*((nk + 31)/32);
61-
int diff = nk32 - nk;
62-
first_k = std::max(0, first_k - diff);
63-
last_k = first_k + nk32;
64-
}
65-
return last_k - first_k < nek1;
66-
}
6746
}
6847

6948
// TODO: get the ggml_type enum here without polution
@@ -87,8 +66,7 @@ extern "C" IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float
8766
const void * k, // k matrix. Assumed to be fp16, nq x nk elements
8867
const void * v, // v matrix. Assumed to be fp16, nq x nk elements
8968
const void * mask, // mask. If not null, assumed to be fp16. nq x nk elements
90-
const void * sinks, // attention sinks
91-
const void * bounds, // attention mask bounds
69+
const void * sinks, // mask. If not null, assumed to be fp16. nq x nk elements
9270
float scale, // scale applied before softmax
9371
float softcap, // if > 0, a "soft-cap" operation is applied before softmax
9472
float * qkv, // v*softmax(scale*(k*q))
@@ -102,13 +80,22 @@ extern "C" IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float
10280
int rk3 = neq3/nek3;
10381
int rv3 = neq3/nev3;
10482

105-
bool range_found = false;
106-
if (neq3 == 1 && rk2 > 1 && neq1 == 1 && bounds && nek1 > 32) {
107-
range_found = true;
108-
auto b = (const int32_t *)bounds;
109-
int first_k = b[0];
110-
int last_k = b[1];
111-
if ((last_k - first_k)%32 == 0) { // why is this not better? : if (reduce_k_range(nek1, first_k, last_k)) {
83+
int first_k = 0, last_k = nek1;
84+
if (neq3 == 1 && rk2 > 1 && neq1 == 1 && nek1 > 256) {
85+
// This is a quick hack for SWA models.
86+
// Given that the mask is the same for all layers, ideally we should determinbe the
87+
// cache bounds once, and reuse for the whole graph. But even with this simple hack
88+
// we get non-negligible performance gains for SWA models and long context.
89+
auto umask = (const uint16_t *)mask;
90+
for (; first_k < last_k; ++first_k) {
91+
if (umask[first_k] == 0) break;
92+
}
93+
for (; last_k > first_k; --last_k) {
94+
if (umask[last_k-1] == 0) break;
95+
}
96+
//printf("nek1 = %d, first = %d, last = %d\n", nek1, first, last);
97+
if (last_k - first_k <= 3*nek1/4 && (last_k - first_k)%32 == 0) {
98+
//printf("Reducing from %d to %d\n", nek1, last_k - first_k);
11299
k = (const void *)((const char *)k + first_k*stride_k);
113100
v = (const void *)((const char *)v + first_k*stride_v);
114101
mask = (const void *)((const uint16_t *)mask + first_k);
@@ -118,7 +105,7 @@ extern "C" IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float
118105

119106
int int_type_k = int_type_k_in;
120107
auto work_buffer = work_buffer_in;
121-
if (neq1 >= 8 || (false && rk2 >= 8 && nek2 > 1)) {
108+
if (neq1 >= 8 || (rk2 >= 8 && nek2 > 1)) {
122109
uint64_t row_size = 0;
123110
work_buffer = iqk_repack_k(int_type_k, Dk, nek1, nek2, nek3, stride_k, nbk2, nbk3, k, work_buffer_in, ith, nth, int_type_k, row_size);
124111
if (int_type_k != int_type_k_in) {
@@ -312,25 +299,6 @@ extern "C" IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float
312299
if (counter++ % (nth/ntg) == ith/ntg) {
313300
int iq1 = (ith%ntg)*neq1g;
314301
int this_neq1 = std::min(neq1g, neq1-iq1);
315-
if (bounds && !range_found) {
316-
auto b = (const int32_t *)bounds + 2*iq1;
317-
int kmin = nek1, kmax = 0;
318-
for (int i = 0; i < this_neq1; ++i) {
319-
kmin = std::min(kmin, b[2*i+0]);
320-
kmax = std::max(kmax, b[2*i+1]);
321-
}
322-
if (reduce_k_range(nek1, kmin, kmax)) {
323-
if (!iqk_flash_attn_impl(int_type_k, int_type_v,
324-
Dk, Dv, this_neq1, kmax-kmin, stride_q, stride_k, stride_v, stride_m, ne1*nb1/sizeof(float),
325-
(const float *)((const char *)q + iq2*nbq2 + iq3*nbq3 + iq1*stride_q),
326-
(const void *)((const char *)k + iq2/rk2*nbk2 + iq3/rk3*nbk3 + kmin*stride_k),
327-
(const void *)((const char *)v + iq2/rv2*nbv2 + iq3/rv3*nbv3 + kmin*stride_v),
328-
(const void *)((const char *)mask + iq1*stride_m + kmin*sizeof(uint16_t)), sinksf, 1,
329-
scale, softcap,
330-
(float *)((char *)qkv + (iq3*ne2*ne1 + iq2 + iq1*ne1)*nb1), nullptr, nullptr)) return false;
331-
continue;
332-
}
333-
}
334302
if (!iqk_flash_attn_impl(int_type_k, int_type_v,
335303
Dk, Dv, this_neq1, nek1, stride_q, stride_k, stride_v, stride_m, ne1*nb1/sizeof(float),
336304
(const float *)((const char *)q + iq2*nbq2 + iq3*nbq3 + iq1*stride_q),

‎ggml/src/iqk/iqk_mul_mat.h‎

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -58,8 +58,7 @@ IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float max_bias,
5858
const void * k, // k matrix. Assumed to be fp16, nq x nk elements
5959
const void * v, // v matrix. Assumed to be fp16, nq x nk elements
6060
const void * mask, // mask. If not null, assumed to be fp16. nq x nk elements
61-
const void * sinks, // attention sinks
62-
const void * bounds, // attention mask bounds
61+
const void * sinks, // mask. If not null, assumed to be fp16. nq x nk elements
6362
float scale, // scale applied before softmax
6463
float softcap, // if > 0, a "soft-cap" operation is applied before softmax
6564
float * qkv, // v*softmax(scale*(k*q))

0 commit comments

Comments
 (0)