ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

sync : ggml

Browse source
This commit is contained in:
Georgi Gerganov 2024-08-27 22:01:45 +03:00
parent 3246fe84d7
commit 231cff5f6f
21 changed files with 1422 additions and 178 deletions
Download patch file Download diff file Expand all files Collapse all files

21
ggml/src/ggml-cuda.cu
View file

@ -9,8 +9,10 @@
#include "ggml-cuda/binbcast.cuh"
#include "ggml-cuda/clamp.cuh"
#include "ggml-cuda/concat.cuh"
#include "ggml-cuda/conv-transpose-1d.cuh"
#include "ggml-cuda/convert.cuh"
#include "ggml-cuda/cpy.cuh"
#include "ggml-cuda/cross-entropy-loss.cuh"
#include "ggml-cuda/diagmask.cuh"
#include "ggml-cuda/dmmv.cuh"
#include "ggml-cuda/fattn.cuh"
@ -29,7 +31,6 @@
#include "ggml-cuda/tsembd.cuh"
#include "ggml-cuda/unary.cuh"
#include "ggml-cuda/upscale.cuh"
#include "ggml-cuda/conv-transpose-1d.cuh"
#include <algorithm>
#include <array>
@ -2181,6 +2182,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
case GGML_OP_ADD:
ggml_cuda_op_add(ctx, dst);
break;
case GGML_OP_SUB:
ggml_cuda_op_sub(ctx, dst);
break;
case GGML_OP_ACC:
ggml_cuda_op_acc(ctx, dst);
break;
@ -2267,6 +2271,12 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
case GGML_OP_SQRT:
ggml_cuda_op_sqrt(ctx, dst);
break;
case GGML_OP_SIN:
ggml_cuda_op_sin(ctx, dst);
break;
case GGML_OP_COS:
ggml_cuda_op_cos(ctx, dst);
break;
case GGML_OP_CLAMP:
ggml_cuda_op_clamp(ctx, dst);
break;
@ -2303,6 +2313,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
case GGML_OP_FLASH_ATTN_EXT:
ggml_cuda_flash_attn_ext(ctx, dst);
break;
case GGML_OP_CROSS_ENTROPY_LOSS:
ggml_cuda_cross_entropy_loss(ctx, dst);
break;
default:
return false;
}
@ -2610,6 +2623,7 @@ GGML_CALL static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t
assert(node->buffer->buft == ggml_backend_cuda_buffer_type(cuda_ctx->device));
for (int j = 0; j < GGML_MAX_SRC; j++) {
if (node->src[j] != nullptr) {
assert(node->src[j]->buffer);
assert(node->src[j]->buffer->buft == ggml_backend_cuda_buffer_type(cuda_ctx->device) || ggml_backend_buffer_is_cuda_split(node->src[j]->buffer));
}
}
@ -2853,12 +2867,15 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
case GGML_OP_TRANSPOSE:
case GGML_OP_NORM:
case GGML_OP_ADD:
case GGML_OP_SUB:
case GGML_OP_MUL:
case GGML_OP_DIV:
case GGML_OP_RMS_NORM:
case GGML_OP_SCALE:
case GGML_OP_SQR:
case GGML_OP_SQRT:
case GGML_OP_SIN:
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_CONT:
case GGML_OP_DIAG_MASK_INF:
@ -2890,6 +2907,8 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
}
return ggml_cuda_info().devices[cuda_ctx->device].cc >= CC_VOLTA &&
op->src[1]->type == GGML_TYPE_F16 && op->src[2]->type == GGML_TYPE_F16;
case GGML_OP_CROSS_ENTROPY_LOSS:
return true;
#endif // defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
default:
return false;