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

Fix Q5_0 alignment issues.

Browse source
This commit is contained in:
Henri Vasserman 2023-05-13 03:37:28 +03:00
parent 3243b9943a
commit cc798cc08c
No known key found for this signature in database
GPG key ID: 2995FC0F58B1A986
1 changed files with 13 additions and 42 deletions
Download patch file Download diff file Expand all files Collapse all files

55
ggml-opencl.c
View file

@ -20,7 +20,7 @@ constant uint QK4_0 = 32;
struct block_q4_0
{
float d;
uchar qs[QK4_0 / 2];
uint8_t qs[QK4_0 / 2];
};
constant uint QK4_1 = 32;
@ -28,31 +28,31 @@ struct block_q4_1
{
float d;
float m;
uchar qs[QK4_1 / 2];
uint8_t qs[QK4_1 / 2];
};
constant uint QK5_0 = 32;
struct block_q5_0
struct __attribute__ ((packed)) block_q5_0
{
float d;
uint qh;
uchar qs[QK5_0 / 2];
half d;
uint32_t qh;
uint8_t qs[QK5_0 / 2];
};
constant uint QK5_1 = 32;
struct block_q5_1
{
ushort d;
ushort m;
uint qh;
uchar qs[QK5_1 / 2];
half d;
half m;
uint32_t qh;
uint8_t qs[QK5_1 / 2];
};
constant uint QK8_0 = 32;
struct block_q8_0
{
float d;
char qs[QK8_0];
uint8_t qs[QK8_0];
};
@ -93,7 +93,7 @@ __kernel void dequantize_row_q5_0(__global struct block_q5_0* x, __global float*
const uint i = get_global_id(0) / qk;
const uint j = get_local_id(0);
const float d = x[i].d;
const float d = vload_half(0, (__global half*) &x[i].d);
uint32_t qh = x[i].qh;
@ -148,20 +148,6 @@ __kernel void dequantize_row_q8_0(__global struct block_q8_0* x, __global float*
} \
} while (0)
#define QK5_0 32
typedef struct {
ggml_fp16_t d; // delta
uint8_t qh[4]; // 5-th bit of quants
uint8_t qs[QK5_0 / 2]; // nibbles / quants
} block_q5_0;
typedef struct {
float d; // delta
uint32_t qh; // 5-th bit of quants
uint8_t qs[QK5_0 / 2]; // nibbles / quants
} cl_block_q5_0;
static cl_platform_id platform;
static cl_device_id device;
static cl_context context;
@ -272,7 +258,6 @@ void ggml_cl_sgemm_wrapper(
cl_kernel kernel;
size_t global = n * k, local, size_qb;
bool dequant;
cl_block_q5_0* cl_host_b;
switch (btype) {
case GGML_TYPE_F32:
@ -294,18 +279,7 @@ void ggml_cl_sgemm_wrapper(
dequant = true;
kernel = kernel_q5_0;
local = 16;
// For some reason OpenCL seems to be incapable of working with structs of size 22.
// 20 and 24 bytes are fine. Workaround to do the fp16 to fp32 step on CPU...
// TODO Find the reason, fix and remove workaround.
const block_q5_0* b = (const block_q5_0*) host_b;
cl_host_b = (cl_block_q5_0*) malloc(sizeof(cl_block_q5_0) * global / 32);
for (size_t i = 0; i < global / 32; i++) {
cl_host_b[i].d = ggml_fp16_to_fp32(b[i].d);
memcpy(&cl_host_b[i].qh, b[i].qh, sizeof(uint32_t));
memcpy(&cl_host_b[i].qs, b[i].qs, QK5_0 / 2);
}
host_b = (const float*) cl_host_b;
size_qb = global * (sizeof(float) + sizeof(uint32_t) + local) / 32;
size_qb = global * (sizeof(ggml_fp16_t) + sizeof(uint32_t) + local) / 32;
break;
case GGML_TYPE_Q5_1:
dequant = true;
@ -384,7 +358,4 @@ void ggml_cl_sgemm_wrapper(
clWaitForEvents(1, &ev_c);
clReleaseEvent(ev_sgemm);
clReleaseEvent(ev_c);
if (btype == GGML_TYPE_Q5_0) {
free((void*) cl_host_b);
}
}