Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

Commit 939ad2d

Browse filesBrowse files
swsw
authored
Fix undefined variables in debug build, remove unused variables (abetlen#531)
1 parent 8c2ec5e commit 939ad2d
Copy full SHA for 939ad2d

File tree

Expand file treeCollapse file tree

1 file changed

+19
-28
lines changed
Filter options
Expand file treeCollapse file tree

1 file changed

+19
-28
lines changed

‎ggml.c

Copy file name to clipboardExpand all lines: ggml.c
+19-28Lines changed: 19 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -1698,8 +1698,6 @@ inline static void ggml_vec_dot_q4_0(const int n, float * restrict s, const void
16981698
// Horizontal sum of all lanes of the accumulator
16991699
sumf = _mm512_reduce_add_ps( acc0 ) + _mm512_reduce_add_ps( acc1 );
17001700
#elif defined(__AVX2__)
1701-
const size_t countBlocks = nb;
1702-
17031701
// Initialize accumulator with zeros
17041702
__m256 acc = _mm256_setzero_ps();
17051703

@@ -5806,23 +5804,28 @@ static void ggml_compute_forward_mul_mat_f32(
58065804
const int ne02 = src0->ne[2];
58075805
const int ne03 = src0->ne[3];
58085806

5807+
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
58095808
const int ne10 = src1->ne[0];
5809+
#endif
58105810
const int ne11 = src1->ne[1];
5811-
//const int ne12 = src1->ne[2];
5812-
//const int ne13 = src1->ne[3];
5811+
#ifndef NDEBUG
5812+
const int ne12 = src1->ne[2];
5813+
const int ne13 = src1->ne[3];
58135814

5814-
//const int ne0 = dst->ne[0];
5815-
//const int ne1 = dst->ne[1];
5816-
//const int ne2 = dst->ne[2];
5817-
//const int ne3 = dst->ne[3];
5818-
//const int ne = ne0*ne1*ne2*ne3;
5815+
const int ne0 = dst->ne[0];
5816+
const int ne1 = dst->ne[1];
5817+
const int ne2 = dst->ne[2];
5818+
const int ne3 = dst->ne[3];
58195819

5820-
//const int nb00 = src0->nb[0];
5820+
const int nb00 = src0->nb[0];
5821+
#endif
58215822
const int nb01 = src0->nb[1];
58225823
const int nb02 = src0->nb[2];
58235824
const int nb03 = src0->nb[3];
58245825

5826+
#ifndef NDEBUG
58255827
const int nb10 = src1->nb[0];
5828+
#endif
58265829
const int nb11 = src1->nb[1];
58275830
const int nb12 = src1->nb[2];
58285831
const int nb13 = src1->nb[3];
@@ -5840,8 +5843,9 @@ static void ggml_compute_forward_mul_mat_f32(
58405843
assert(ne2 == ne12);
58415844
assert(ne3 == ne13);
58425845

5843-
// TODO: we don't support permuted src0
5846+
// we don't support permuted src0 or src1
58445847
assert(nb00 == sizeof(float));
5848+
assert(nb10 == sizeof(float));
58455849

58465850
// dst cannot be transposed or permuted
58475851
assert(nb0 == sizeof(float));
@@ -5859,8 +5863,6 @@ static void ggml_compute_forward_mul_mat_f32(
58595863

58605864
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
58615865
if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
5862-
GGML_ASSERT(nb10 == sizeof(float));
5863-
58645866
if (params->ith != 0) {
58655867
return;
58665868
}
@@ -5903,9 +5905,6 @@ static void ggml_compute_forward_mul_mat_f32(
59035905
return;
59045906
}
59055907

5906-
// TODO: do not support transposed src1
5907-
assert(nb10 == sizeof(float));
5908-
59095908
// parallelize by src0 rows using ggml_vec_dot_f32
59105909

59115910
// total rows in src0
@@ -6169,7 +6168,6 @@ static void ggml_compute_forward_mul_mat_q4_0_f32(
61696168
const int ne1 = dst->ne[1];
61706169
const int ne2 = dst->ne[2];
61716170
const int ne3 = dst->ne[3];
6172-
//const int ne = ne0*ne1*ne2*ne3;
61736171

61746172
const int nb00 = src0->nb[0];
61756173
const int nb01 = src0->nb[1];
@@ -6194,8 +6192,9 @@ static void ggml_compute_forward_mul_mat_q4_0_f32(
61946192
GGML_ASSERT(ne2 == ne12);
61956193
GGML_ASSERT(ne3 == ne13);
61966194

6197-
// TODO: we don't support permuted src0
6195+
// we don't support permuted src0 or src1
61986196
GGML_ASSERT(nb00 == (int) GGML_TYPE_SIZE[GGML_TYPE_Q4_0]);
6197+
GGML_ASSERT(nb10 == sizeof(float));
61996198

62006199
// dst cannot be transposed or permuted
62016200
GGML_ASSERT(nb0 == sizeof(float));
@@ -6213,8 +6212,6 @@ static void ggml_compute_forward_mul_mat_q4_0_f32(
62136212

62146213
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
62156214
if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
6216-
GGML_ASSERT(nb10 == sizeof(float));
6217-
62186215
if (params->ith != 0) {
62196216
return;
62206217
}
@@ -6278,8 +6275,6 @@ static void ggml_compute_forward_mul_mat_q4_0_f32(
62786275
return;
62796276
}
62806277

6281-
// TODO: do not support transposed src1
6282-
62836278
// parallelize by src0 rows using ggml_vec_dot_q4_0
62846279

62856280
// total rows in src0
@@ -6354,7 +6349,6 @@ static void ggml_compute_forward_mul_mat_q4_1_f32(
63546349
const int ne1 = dst->ne[1];
63556350
const int ne2 = dst->ne[2];
63566351
const int ne3 = dst->ne[3];
6357-
//const int ne = ne0*ne1*ne2*ne3;
63586352

63596353
const int nb00 = src0->nb[0];
63606354
const int nb01 = src0->nb[1];
@@ -6379,8 +6373,9 @@ static void ggml_compute_forward_mul_mat_q4_1_f32(
63796373
GGML_ASSERT(ne2 == ne12);
63806374
GGML_ASSERT(ne3 == ne13);
63816375

6382-
// TODO: we don't support permuted src0
6376+
// we don't support permuted src0 or src1
63836377
GGML_ASSERT(nb00 == (int) GGML_TYPE_SIZE[GGML_TYPE_Q4_1]);
6378+
GGML_ASSERT(nb10 == sizeof(float));
63846379

63856380
// dst cannot be transposed or permuted
63866381
GGML_ASSERT(nb0 == sizeof(float));
@@ -6398,8 +6393,6 @@ static void ggml_compute_forward_mul_mat_q4_1_f32(
63986393

63996394
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
64006395
if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
6401-
GGML_ASSERT(nb10 == sizeof(float));
6402-
64036396
if (params->ith != 0) {
64046397
return;
64056398
}
@@ -6466,8 +6459,6 @@ static void ggml_compute_forward_mul_mat_q4_1_f32(
64666459
return;
64676460
}
64686461

6469-
// TODO: do not support transposed src1
6470-
64716462
// parallelize by src0 rows using ggml_vec_dot_q4_1
64726463

64736464
// total rows in src0

0 commit comments

Comments
0 (0)
Morty Proxy This is a proxified and sanitized view of the page, visit original site.