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

metal : fix API debug warnings

Browse source
This commit is contained in:
Georgi Gerganov 2023-12-30 21:10:32 +02:00
parent a184e1050c
commit 1580805fc6
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
2 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-metal.m
View file

@ -1238,7 +1238,7 @@ void ggml_metal_graph_compute(
// not sure how to avoid this // not sure how to avoid this
// TODO: make a simpler cpy_bytes kernel // TODO: make a simpler cpy_bytes kernel
const int nth = MIN(1024, ne00); const int nth = MIN((int) ctx->pipeline_cpy_f32_f32.maxTotalThreadsPerThreadgroup, ne00);
[encoder setComputePipelineState:ctx->pipeline_cpy_f32_f32]; [encoder setComputePipelineState:ctx->pipeline_cpy_f32_f32];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0]; [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
@ -2239,7 +2239,7 @@ void ggml_metal_graph_compute(
[encoder setBytes:&nb3 length:sizeof(nb3) atIndex:17]; [encoder setBytes:&nb3 length:sizeof(nb3) atIndex:17];
[encoder setBytes:&sf length:sizeof(sf) atIndex:18]; [encoder setBytes:&sf length:sizeof(sf) atIndex:18];
const int nth = MIN(1024, ne0); const int nth = MIN((int) ctx->pipeline_upscale_f32.maxTotalThreadsPerThreadgroup, ne0);
[encoder dispatchThreadgroups:MTLSizeMake(ne1, ne2, ne3) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)]; [encoder dispatchThreadgroups:MTLSizeMake(ne1, ne2, ne3) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
} break; } break;

2
ggml-metal.metal
View file

@ -3523,7 +3523,7 @@ void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg
device const int8_t * qs = ((device const int8_t *)xb->qs); device const int8_t * qs = ((device const int8_t *)xb->qs);
const half d = xb->d; const half d = xb->d;
for (int i=0;i<16;i++) { for (int i = 0; i < 16; i++) {
reg[i/4][i%4] = (qs[i + 16*il] * d); reg[i/4][i%4] = (qs[i + 16*il] * d);
} }
} }