vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

ggml : reduce hash table reset cost (#8698)

Browse source 
* ggml : reduce hash table reset cost

* fix unreachable code warnings after GGML_ASSERT(false)

* GGML_ASSERT(false) -> GGML_ABORT("fatal error")

* GGML_ABORT use format string
This commit is contained in:
slaren 2024-07-27 04:41:55 +02:00 • committed by GitHub
parent 01245f5b16
commit 2b1f616b20
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
46 changed files with 851 additions and 754 deletions
Download patch file Download diff file Expand all files Collapse all files

22
ggml/src/ggml-sycl/mmq.cpp
View file

@ -1799,7 +1799,7 @@ static void ggml_mul_mat_q4_0_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q4_0_PASCAL;
nwarps = NWARPS_Q4_0_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -1914,7 +1914,7 @@ static void ggml_mul_mat_q4_1_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q4_1_PASCAL;
nwarps = NWARPS_Q4_1_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -2029,7 +2029,7 @@ static void ggml_mul_mat_q5_0_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q5_0_PASCAL;
nwarps = NWARPS_Q5_0_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -2144,7 +2144,7 @@ static void ggml_mul_mat_q5_1_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q5_1_PASCAL;
nwarps = NWARPS_Q5_1_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -2259,7 +2259,7 @@ static void ggml_mul_mat_q8_0_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q8_0_PASCAL;
nwarps = NWARPS_Q8_0_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -2374,7 +2374,7 @@ static void ggml_mul_mat_q2_K_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q2_K_PASCAL;
nwarps = NWARPS_Q2_K_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -2497,7 +2497,7 @@ static void ggml_mul_mat_q3_K_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q3_K_PASCAL;
nwarps = NWARPS_Q3_K_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -2625,7 +2625,7 @@ static void ggml_mul_mat_q4_K_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q4_K_PASCAL;
nwarps = NWARPS_Q4_K_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -2746,7 +2746,7 @@ static void ggml_mul_mat_q5_K_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q5_K_PASCAL;
nwarps = NWARPS_Q5_K_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -2867,7 +2867,7 @@ static void ggml_mul_mat_q6_K_q8_1_sycl(const void *vx, const void *vy,
mmq_y = MMQ_Y_Q6_K_PASCAL;
nwarps = NWARPS_Q6_K_PASCAL;
} else {
GGML_ASSERT(false);
GGML_ABORT("fatal error");
}
const int block_num_x = (nrows_x + mmq_y - 1) / mmq_y;
@ -3016,7 +3016,7 @@ void ggml_sycl_op_mul_mat_q(
ggml_mul_mat_q6_K_q8_1_sycl(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, src1_ncols, src1_padded_row_size, nrows_dst, stream);
break;
default:
GGML_ASSERT(false);
GGML_ABORT("fatal error");
break;
}