Skip to content

Instantly share code, notes, and snippets.

@ochafik

ochafik/ggml-opencl.cl Secret

Last active December 6, 2023 23:31
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save ochafik/aeda893dec73c2181a88752022f129e5 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save ochafik/aeda893dec73c2181a88752022f129e5 to your computer and use it in GitHub Desktop.
Download ZIP
llama.cpp + RPi OpenCL (clvk --> Vulkan)
Raw
ggml-opencl.cl
/**
TODO
- Checkin unmodified cl file
- Update all const int to const size_t (w/ modulos if needed)
- Polyfill for cl_khr_fp16 (vload_half, half defines) tested on MacOS X
- https://gist.github.com/milhidaka/95863906fe828198f47991c813dbe233
- https://github.com/AcademySoftwareFoundation/Imath/blob/490be7e74e713b119895547a9d78a616ed479ceb/src/Imath/half.h#L285
- https://stackoverflow.com/questions/1659440/32-bit-to-16-bit-floating-point-conversion/60047308#60047308
- FAST MATH detection for faster conversion!
- Functions to get all block_* fields out of array of uint32_t (for RPi), and turn vload_half into HALF_TO_FLOAT(uint32_t)
- Bump https://github.com/ggerganov/llama.cpp/issues/1571
- File bug for clspv: '__kernel void bug(__global ushort* x, __global float* y) { y[get_global_id(0)] = x[get_global_id(0)]; }'
*/
// #define CLSPV_WORKAROUNDS 1
// #ifdef cl_khr_fp16
// #pragma OPENCL EXTENSION cl_khr_fp16 : enable
// #else
// #define half uint16_t
// #endif // cl_khr_fp16
typedef char int8_t;
typedef uchar uint8_t;
typedef short int16_t;
typedef ushort uint16_t;
typedef int int32_t;
typedef uint uint32_t;
// https://github.com/AcademySoftwareFoundation/Imath/blob/490be7e74e713b119895547a9d78a616ed479ceb/src/Imath/half.h#L285
// https://stackoverflow.com/questions/1659440/32-bit-to-16-bit-floating-point-conversion/60047308#60047308
static float half_to_float(uint32_t h) {
union { uint32_t i; float f; } v;
uint32_t hexpmant = ((uint32_t) (h) << 17) >> 4;
v.i = ((uint32_t) (h >> 15)) << 31;
if (hexpmant >= 0x00800000)
{
v.i |= hexpmant;
// either we are a normal number, in which case add in the bias difference
// otherwise make sure all exponent bits are set
if (hexpmant < 0x0f800000)
v.i += 0x38000000;
else
v.i |= 0x7f800000;
}
else if (hexpmant != 0)
{
// exponent is 0 because we're denormal, don't have to extract
// the mantissa, can just use as is
uint32_t lc = clz(hexpmant) - 8;
// so nominally we want to remove that extra bit we shifted
// up, but we are going to add that bit back in, then subtract
// from it with the 0x38800000 - (lc << 23)....
//
// by combining, this allows us to skip the & operation (and
// remove a constant)
//
// hexpmant &= ~0x00800000;
v.i |= 0x38800000;
// lc is now x, where the desired exponent is then
// -14 - lc
// + 127 -> new exponent
v.i |= (hexpmant << lc);
v.i -= (lc << 23);
}
// if (true) return 12.3f;
return v.f;
}
#ifdef CLSPV_WORKAROUNDS
#ifdef __ENDIAN_LITTLE__
#define UCHAR_FROM_UINT(x, offset) ((offset) == 0 ? ((x) & 0xFF) : (offset) == 1 ? (((x) >> 8) & 0xFF) : (offset) == 2 ? (((x) >> 16) & 0xFF) : (((x) >> 24)))
#define USHORT_FROM_UINT(x, offset) ((offset) == 0 ? ((x) & 0xFFFF) : (offset) == 1 ? (((x) >> 8) & 0xFFFF) : (offset) == 2 ? ((x) >> 16) : 0)
#define FIRST_BITS_FROM_UINT(x, bits) ((x) & ((1 << (bits)) - 1))
#define LAST_BITS_FROM_UINT(x, bits) (((x) >> (bits)) & ((1 << (bits)) - 1))
#define CONCAT_BITS(x, xbits, y, ybits) ((x) | ((y) << (ybits)))
#else
#define UCHAR_FROM_UINT(x, offset) ((offset) == 0 ? ((x >> 24)) : (offset) == 1 ? ((x >> 16) & 0xFF) : (offset) == 2 ? ((x >> 8) & 0xFF) : ((x) & 0xFF))
#define USHORT_FROM_UINT(x, offset) ((offset) == 0 ? ((x) >> 16) : (offset) == 1 ? (((x) >> 8) & 0xFFFF) : (offset) == 2 ? ((x) & 0xFFFF) : 0)
#define LAST_BITS_FROM_UINT(x, bits) ((x) & ((1 << (bits)) - 1))
#define FIRST_BITS_FROM_UINT(x, bits) (((x) >> (bits)) & ((1 << (bits)) - 1))
#define CONCAT_BITS(x, xbits, y, ybits) ((x) << (xbits) | (y))
#endif
static uint32_t get_uint32(__global const uint32_t* x, size_t offset) {
const uint8_t m = offset % 4;
const size_t i = offset / 4;
const uint32_t x0 = x[i];
if (m == 0) {
return x0;
}
const uint32_t x1 = x[i + 1];
const uint8_t b1 = m * 8;
const uint8_t b0 = 32 - b1;
return CONCAT_BITS(LAST_BITS_FROM_UINT(x0, b0), b0, FIRST_BITS_FROM_UINT(x1, b1), b1);
}
static uint16_t get_uint16(__global const uint32_t* x, size_t offset) {
const uint8_t m = offset % 4;
const size_t i = offset / 4;
const uint32_t x0 = x[i];
if (m < 3) {
return USHORT_FROM_UINT(x0, m);
}
const uint32_t x1 = x[i + 1];
return CONCAT_BITS(LAST_BITS_FROM_UINT(x0, 8), 8, FIRST_BITS_FROM_UINT(x1, 8), 8);
}
static float get_half(__global const uint32_t* x, size_t offset) {
return half_to_float(get_uint16(x, offset));
}
static uint8_t get_uint8(__global const uint32_t* x, size_t offset) {
const uint8_t m = offset % 4;
const size_t i = offset / 4;
const uint32_t x0 = x[i];
return UCHAR_FROM_UINT(x0, m);
}
static int8_t get_int8(__global const uint32_t* x, size_t offset) {
return (int8_t)get_uint8(x, offset);
}
#endif // CLSPV_WORKAROUNDS
struct __attribute__ ((packed)) block_q4_0
{
half d;
uint8_t qs[QK4_0 / 2];
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q4_0_D(p, i) vload_half(0, &p[i].d)
#define BLOCK_Q4_0_QS(p, i, iQ) p[i].qs[iQ]
#else
#define BLOCK_Q4_0_D(p, i) get_half (p, (i) * sizeof(struct block_q4_0))
#define BLOCK_Q4_0_QS(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q4_0) + 2 + (iQ))
#endif
struct __attribute__ ((packed)) block_q4_1
{
half d;
half m;
uint8_t qs[QK4_1 / 2];
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q4_1_D(p, i) vload_half(0, &p[i].d)
#define BLOCK_Q4_1_M(p, i) vload_half(0, &p[i].m)
#define BLOCK_Q4_1_QS(p, i, iQ) p[i].qs[iQ]
#else
#define BLOCK_Q4_1_D(p, i) get_half (p, (i) * sizeof(struct block_q4_1))
#define BLOCK_Q4_1_M(p, i) get_half (p, (i) * sizeof(struct block_q4_1) + 2)
#define BLOCK_Q4_1_QS(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q4_1) + 4 + (iQ))
#endif
struct __attribute__ ((packed)) block_q5_0
{
half d;
uint32_t qh;
uint8_t qs[QK5_0 / 2];
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q5_0_D(p, i) vload_half(0, &p[i].d)
#define BLOCK_Q5_0_QH(p, i) p[i].qh
#define BLOCK_Q5_0_QS(p, i, iQ) p[i].qs[iQ]
#else
#define BLOCK_Q5_0_D(p, i) get_half (p, (i) * sizeof(struct block_q5_0))
#define BLOCK_Q5_0_QH(p, i) get_uint32(p, (i) * sizeof(struct block_q5_0) + 2)
#define BLOCK_Q5_0_QS(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q5_0) + 2 + 4 + (iQ))
#endif
struct __attribute__ ((packed)) block_q5_1
{
half d;
half m;
uint32_t qh;
uint8_t qs[QK5_1 / 2];
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q5_1_D(p, i) vload_half(0, &p[i].d)
#define BLOCK_Q5_1_M(p, i) vload_half(0, &p[i].m)
#define BLOCK_Q5_1_QH(p, i) p[i].qh
#define BLOCK_Q5_1_QS(p, i, iQ) p[i].qs[iQ]
#else
#define BLOCK_Q5_1_D(p, i) get_half (p, (i) * sizeof(struct block_q5_1))
#define BLOCK_Q5_1_M(p, i) get_half (p, (i) * sizeof(struct block_q5_1) + 2)
#define BLOCK_Q5_1_QH(p, i) get_uint32(p, (i) * sizeof(struct block_q5_1) + 2 + 2)
#define BLOCK_Q5_1_QS(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q5_1) + 2 + 2 + 4, (iQ))
#endif
struct __attribute__ ((packed)) block_q8_0
{
half d;
int8_t qs[QK8_0];
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q8_0_D(p, i) vload_half(0, &p[i].d)
#define BLOCK_Q8_0_QS(p, i, iQ) p[i].qs[iQ]
#else
#define BLOCK_Q8_0_D(p, i) get_half (p, (i) * sizeof(struct block_q8_0))
#define BLOCK_Q8_0_QS(p, i, iQ) get_int8 (p, (i) * sizeof(struct block_q8_0) + 2 + (iQ))
#endif
struct __attribute__((packed)) block_q2_K
{
uint8_t scales[16];
uint8_t qs[64];
half d;
half dmin;
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q2_K_SCALES(p, i, iScale) p[i].scales[iScale]
#define BLOCK_Q2_K_QS(p, i, iQ) p[i].qs[iQ]
#define BLOCK_Q2_K_D(p, i) vload_half(0, &p[i].d)
#define BLOCK_Q2_K_DMIN(p, i) vload_half(0, &p[i].dmin)
#else
#define BLOCK_Q2_K_SCALES(p, i, iScale) get_uint8 (p, (i) * sizeof(struct block_q2_K) + (iScale))
#define BLOCK_Q2_K_QS(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q2_K) + 16 + (iQ))
#define BLOCK_Q2_K_D(p, i) get_half (p, (i) * sizeof(struct block_q2_K) + 16 + 64)
#define BLOCK_Q2_K_DMIN(p, i) get_half (p, (i) * sizeof(struct block_q2_K) + 16 + 64 + 2)
#endif
struct __attribute__((packed)) block_q3_K
{
uint8_t hmask[32];
uint8_t qs[64];
uint8_t scales[12];
half d;
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q3_K_HMASK(p, i, iM) p[i].hmask[iM]
#define BLOCK_Q3_K_QS(p, i, iQ) p[i].qs[iQ]
#define BLOCK_Q3_K_SCALES(p, i, iScale) p[i].scales[iScale]
#define BLOCK_Q3_K_D(p, i) vload_half(0, &p[i].d)
#else
#define BLOCK_Q3_K_HMASK(p, i, iM) get_uint8 (p, (i) * sizeof(struct block_q3_K) + iM)
#define BLOCK_Q3_K_QS(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q3_K) + 32 + (iQ))
#define BLOCK_Q3_K_SCALES(p, i, iScale) get_uint8 (p, (i) * sizeof(struct block_q3_K) + 32 + 64 + (iScale))
#define BLOCK_Q3_K_D(p, i) get_half (p, (i) * sizeof(struct block_q3_K) + 32 + 64 + 12)
#endif
struct __attribute__((packed)) block_q4_K
{
half d;
half dmin;
uint8_t scales[12];
uint8_t qs[128];
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q4_K_D(p, i) vload_half(0, &p[i].d)
#define BLOCK_Q4_K_DMIN(p, i) vload_half(0, &p[i].dmin)
#define BLOCK_Q4_K_SCALES(p, i, iScale) p[i].scales[iScale]
#define BLOCK_Q4_K_QS(p, i, iQ) p[i].qs[iQ]
#else
#define BLOCK_Q4_K_D(p, i) get_half (p, (i) * sizeof(struct block_q4_K))
#define BLOCK_Q4_K_DMIN(p, i) get_half (p, (i) * sizeof(struct block_q4_K) + 2)
#define BLOCK_Q4_K_SCALES(p, i, iScale) get_uint8 (p, (i) * sizeof(struct block_q4_K) + 2 + 2 + (iScale))
#define BLOCK_Q4_K_QS(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q4_K) + 2 + 2 + 12 + (iQ))
#endif
struct __attribute__((packed)) block_q5_K
{
half d;
half dmin;
uint8_t scales[12];
uint8_t qh[32];
uint8_t qs[128];
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q5_K_D(p, i) vload_half(0, &p[i].d)
#define BLOCK_Q5_K_DMIN(p, i) vload_half(0, &p[i].dmin)
#define BLOCK_Q5_K_SCALES(p, i, iScale) p[i].scales[iScale]
#define BLOCK_Q5_K_QH(p, i, iQ) p[i].qh[iQ]
#define BLOCK_Q5_K_QS(p, i, iQ) p[i].qs[iQ]
#else
#define BLOCK_Q5_K_D(p, i) get_half (p, (i) * sizeof(struct block_q5_K))
#define BLOCK_Q5_K_DMIN(p, i) get_half (p, (i) * sizeof(struct block_q5_K) + 2)
#define BLOCK_Q5_K_SCALES(p, i, iScale) get_uint8 (p, (i) * sizeof(struct block_q5_K) + 2 + 2 + (iScale))
#define BLOCK_Q5_K_QH(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q5_K) + 2 + 2 + 12 + (iQ))
#define BLOCK_Q5_K_QS(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q5_K) + 2 + 2 + 12 + 32 + (iQ))
#endif
struct __attribute__((packed)) block_q6_K
{
uint8_t ql[128];
uint8_t qh[64];
int8_t scales[16];
half d;
};
#ifndef CLSPV_WORKAROUNDS
#define BLOCK_Q6_K_QL(p, i, iQ) p[i].ql[iQ]
#define BLOCK_Q6_K_QH(p, i, iQ) p[i].qh[iQ]
#define BLOCK_Q6_K_SCALES(p, i, iScale) p[i].scales[iScale]
#define BLOCK_Q6_K_D(p, i) vload_half(0, &p[i].d)
#else
#define BLOCK_Q6_K_QL(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q6_K) + (iQ))
#define BLOCK_Q6_K_QH(p, i, iQ) get_uint8 (p, (i) * sizeof(struct block_q6_K) + 128 + (iQ))
#define BLOCK_Q6_K_SCALES(p, i, iScale) get_int8 (p, (i) * sizeof(struct block_q6_K) + 128 + 64 + (iScale))
#define BLOCK_Q6_K_D(p, i) get_half (p, (i) * sizeof(struct block_q6_K) + 128 + 64 + 16)
#endif
#ifndef CLSPV_WORKAROUNDS
__kernel void convert_fp16_to_fp32(__global half* x, __global float* y) {
const uint i = get_global_id(0);
y[i] = vload_half(0, &x[i]);
}
#else
__kernel void convert_fp16_to_fp32(__global uint32_t* x, __global float* y) {
const uint i = get_global_id(0);
y[i] = get_half(x, i * 2);
}
#endif
static void dequantize_q4_0(__global const struct block_q4_0* x, const int ib, const int iqs, float* v0, float* v1) {
const float d = BLOCK_Q4_0_D(x, ib);
const uint8_t vui = BLOCK_Q4_0_QS(x, ib, iqs);
const int8_t vi0 = vui & 0xF;
const int8_t vi1 = vui >> 4;
*v0 = (vi0 - 8)*d;
*v1 = (vi1 - 8)*d;
}
static void dequantize_q4_1(__global const struct block_q4_1* x, const int ib, const int iqs, float* v0, float* v1) {
const float d = BLOCK_Q4_1_D(x, ib);
const float m = BLOCK_Q4_1_M(x, ib);
const uint8_t vui = BLOCK_Q4_1_QS(x, ib, iqs);
const int8_t vi0 = vui & 0xF;
const int8_t vi1 = vui >> 4;
*v0 = vi0*d + m;
*v1 = vi1*d + m;
}
static void dequantize_q5_0(__global const struct block_q5_0* x, const int ib, const int iqs, float* v0, float* v1) {
const float d = BLOCK_Q5_0_D(x, ib);
uint32_t qh = BLOCK_Q5_0_QH(x, ib);
const uint8_t xh_0 = ((qh >> (iqs + 0)) << 4) & 0x10;
const uint8_t xh_1 = ((qh >> (iqs + 12)) ) & 0x10;
const int32_t x0 = ((BLOCK_Q5_0_QS(x, ib, iqs) & 0xf) | xh_0) - 16;
const int32_t x1 = ((BLOCK_Q5_0_QS(x, ib, iqs) >> 4) | xh_1) - 16;
*v0 = x0*d;
*v1 = x1*d;
}
static void dequantize_q5_1(__global const struct block_q5_1* x, const int ib, const int iqs, float* v0, float* v1) {
const float d = BLOCK_Q5_1_D(x, ib);
const float m = BLOCK_Q5_1_M(x, ib);
uint32_t qh = BLOCK_Q5_1_QH(x, ib);
const uint8_t xh_0 = ((qh >> (iqs + 0)) << 4) & 0x10;
const uint8_t xh_1 = ((qh >> (iqs + 12)) ) & 0x10;
const int32_t x0 = ((BLOCK_Q5_1_QS(x, ib, iqs) & 0xf) | xh_0);
const int32_t x1 = ((BLOCK_Q5_1_QS(x, ib, iqs) >> 4) | xh_1);
*v0 = x0*d + m;
*v1 = x1*d + m;
}
static void dequantize_q8_0(__global const struct block_q8_0* x, const int ib, const int iqs, float* v0, float* v1) {
const float d = BLOCK_Q8_0_D(x, ib);
const int8_t vi0 = x[ib].qs[iqs + 0];
const int8_t vi1 = x[ib].qs[iqs + 1];
*v0 = vi0*d;
*v1 = vi1*d;
}
#ifndef CLSPV_WORKAROUNDS
inline convert_f16(__global half* x, const int ib, const int iqs, float* v0, float* v1){
*v0 = vload_half(0, &x[ib + 0]);
*v1 = vload_half(0, &x[ib + 1]);
}
#else
inline void convert_f16(__global uint32_t* x, const int ib, const int iqs, float* v0, float* v1){
*v0 = get_half(x, 2 * (ib + 0));
*v1 = get_half(x, 2 * (ib + 1));
}
#endif
inline void get_scale_min_k4(int j, const __global uint8_t *q, uint8_t *d, uint8_t *m)
{
if (j < 4)
{
*d = q[j] & 63;
*m = q[j + 4] & 63;
}
else
{
*d = (q[j + 4] & 0xF) | ((q[j - 4] >> 6) << 4);
*m = (q[j + 4] >> 4) | ((q[j - 0] >> 6) << 4);
}
}
__kernel void dequantize_block_q2_K(__global const struct block_q2_K *x, __global float *yy)
{
const int i = get_group_id(0) + get_global_offset(0);
const int tid = get_local_id(0);
const int n = tid / 32;
const int l = tid - 32 * n;
const int is = 8 * n + l / 16;
const uint8_t q = BLOCK_Q2_K_QS(x, i, 32 * n + l);
__global float *y = yy + get_group_id(0) * QK_K + 128 * n;
const float dall = BLOCK_Q2_K_D(x, i);
const float dmin = BLOCK_Q2_K_DMIN(x, i);
y[l + 0] = dall * (BLOCK_Q2_K_SCALES(x, i, is + 0) & 0xF) * ((q >> 0) & 3) - dmin * (BLOCK_Q2_K_SCALES(x, i, is + 0) >> 4);
y[l + 32] = dall * (BLOCK_Q2_K_SCALES(x, i, is + 2) & 0xF) * ((q >> 2) & 3) - dmin * (BLOCK_Q2_K_SCALES(x, i, is + 2) >> 4);
y[l + 64] = dall * (BLOCK_Q2_K_SCALES(x, i, is + 4) & 0xF) * ((q >> 4) & 3) - dmin * (BLOCK_Q2_K_SCALES(x, i, is + 4) >> 4);
y[l + 96] = dall * (BLOCK_Q2_K_SCALES(x, i, is + 6) & 0xF) * ((q >> 6) & 3) - dmin * (BLOCK_Q2_K_SCALES(x, i, is + 6) >> 4);
}
__kernel void dequantize_block_q3_K(__global const struct block_q3_K *x, __global float *yy)
{
int r = get_local_id(0) / 4;
int i = get_group_id(0) + get_global_offset(0);
int tid = r / 2;
int is0 = r % 2;
int l0 = 16 * is0 + 4 * (get_local_id(0) % 4);
int n = tid / 4;
int j = tid - 4 * n;
uint8_t m = 1 << (4 * n + j);
int is = 8 * n + 2 * j + is0;
int shift = 2 * j;
int8_t us = is < 4 ? (BLOCK_Q3_K_SCALES(x, i, is - 0) & 0xF) | (((BLOCK_Q3_K_SCALES(x, i, is + 8) >> 0) & 3) << 4)
: is < 8 ? (BLOCK_Q3_K_SCALES(x, i, is - 0) & 0xF) | (((BLOCK_Q3_K_SCALES(x, i, is + 4) >> 2) & 3) << 4)
: is < 12 ? (BLOCK_Q3_K_SCALES(x, i, is - 8) >> 4) | (((BLOCK_Q3_K_SCALES(x, i, is + 0) >> 4) & 3) << 4)
: (BLOCK_Q3_K_SCALES(x, i, is - 8) >> 4) | (((BLOCK_Q3_K_SCALES(x, i, is - 4) >> 6) & 3) << 4);
float d_all = BLOCK_Q3_K_D(x, i);
float dl = d_all * (us - 32);
__global float *y = yy + get_group_id(0) * QK_K + 128 * n + 32 * j;
const size_t qs_offset = 32 * n;
for (int l = l0; l < l0 + 4; ++l)
y[l] = dl * ((int8_t)((BLOCK_Q3_K_QS(x, i, qs_offset + l) >> shift) & 3) - ((BLOCK_Q3_K_HMASK(x, i, l) & m) ? 0 : 4));
}
__kernel void dequantize_block_q4_K(__global const struct block_q4_K *x, __global float *yy)
{
const int i = get_group_id(0) + get_global_offset(0);
const int tid = get_local_id(0);
const int il = tid / 8;
const int ir = tid % 8;
const int is = 2 * il;
const int n = 4;
__global float *y = yy + get_group_id(0) * QK_K + 64 * il + n * ir;
const float dall = BLOCK_Q4_K_D(x, i);
const float dmin = BLOCK_Q4_K_DMIN(x, i);
const int qs_offset = 32 * il + n * ir;
uint8_t sc, m;
get_scale_min_k4(is + 0, x[i].scales, &sc, &m);
float d1 = dall * sc;
float m1 = dmin * m;
get_scale_min_k4(is + 1, x[i].scales, &sc, &m);
float d2 = dall * sc;
float m2 = dmin * m;
for (int l = 0; l < n; ++l)
{
uint8_t q = BLOCK_Q4_K_QS(x, i, qs_offset + l);
y[l + 0] = d1 * (q & 0xF) - m1;
y[l + 32] = d2 * (q >> 4) - m2;
}
}
__kernel void dequantize_block_q5_K(__global const struct block_q5_K *x, __global float *yy)
{
const int i = get_group_id(0) + get_global_offset(0);
const int tid = get_local_id(0);
const int il = tid / 16;
const int ir = tid % 16;
const int is = 2 * il;
__global float *y = yy + get_group_id(0) * QK_K + 64 * il + 2 * ir;
const float dall = BLOCK_Q5_K_D(x, i);
const float dmin = BLOCK_Q5_K_DMIN(x, i);
const int ql_offset = 32 * il + 2 * ir;
const int qh_offset = 2 * ir;
uint8_t sc, m;
get_scale_min_k4(is + 0, x[i].scales, &sc, &m);
const float d1 = dall * sc;
const float m1 = dmin * m;
get_scale_min_k4(is + 1, x[i].scales, &sc, &m);
const float d2 = dall * sc;
const float m2 = dmin * m;
uint8_t hm = 1 << (2 * il);
uint8_t ql0 = BLOCK_Q5_K_QS(x, i, ql_offset + 0);
uint8_t ql1 = BLOCK_Q5_K_QS(x, i, ql_offset + 1);
uint8_t qh0 = BLOCK_Q5_K_QH(x, i, qh_offset + 0);
uint8_t qh1 = BLOCK_Q5_K_QH(x, i, qh_offset + 1);
y[0] = d1 * ((ql0 & 0xF) + (qh0 & hm ? 16 : 0)) - m1;
y[1] = d1 * ((ql1 & 0xF) + (qh1 & hm ? 16 : 0)) - m1;
hm <<= 1;
y[32] = d2 * ((ql0 >> 4) + (qh0 & hm ? 16 : 0)) - m2;
y[33] = d2 * ((ql1 >> 4) + (qh1 & hm ? 16 : 0)) - m2;
}
__kernel void dequantize_block_q6_K(__global const struct block_q6_K *x, __global float *yy)
{
const int i = get_group_id(0) + get_global_offset(0);
const int tid = get_local_id(0);
const int ip = tid / 32;
const int il = tid - 32 * ip;
const int is = 8 * ip + il / 16;
__global float *y = yy + get_group_id(0) * QK_K + 128 * ip + il;
const float d = BLOCK_Q6_K_D(x, i);
const int ql_offset = 64 * ip + il;
const uint8_t qh = BLOCK_Q6_K_QH(x, i, 32 * ip + il);
y[0] = d * BLOCK_Q6_K_SCALES(x, i, is + 0) * ((int8_t)((BLOCK_Q6_K_QL(x, i, ql_offset + 0) & 0xF) | (((qh >> 0) & 3) << 4)) - 32);
y[32] = d * BLOCK_Q6_K_SCALES(x, i, is + 2) * ((int8_t)((BLOCK_Q6_K_QL(x, i, ql_offset + 32) & 0xF) | (((qh >> 2) & 3) << 4)) - 32);
y[64] = d * BLOCK_Q6_K_SCALES(x, i, is + 4) * ((int8_t)((BLOCK_Q6_K_QL(x, i, ql_offset + 0) >> 4) | (((qh >> 4) & 3) << 4)) - 32);
y[96] = d * BLOCK_Q6_K_SCALES(x, i, is + 6) * ((int8_t)((BLOCK_Q6_K_QL(x, i, ql_offset + 32) >> 4) | (((qh >> 6) & 3) << 4)) - 32);
}
__kernel void dequantize_mul_mat_vec_q2_K(__global const struct block_q2_K * xx, __local float* tmp, __global float* yy, __global float* dst, const int ncols) {
const int row = get_group_id(0);
const int num_blocks_per_row = ncols / QK_K;
const int ib0 = row*num_blocks_per_row + get_global_offset(0);
__global const struct block_q2_K * x = xx + ib0;
const int tid = get_local_id(0)/K_QUANTS_PER_ITERATION; // 0...31 or 0...15
const int ix = get_local_id(0)%K_QUANTS_PER_ITERATION; // 0 or 0,1
const int step = 16/K_QUANTS_PER_ITERATION;
const int im = tid/step; // 0 or 1. 0 computes 0..., 1 computes 128...
const int in = tid - step*im; // 0...15 or 0...7
const int l0 = K_QUANTS_PER_ITERATION*in; // 0...15 or 0...14 in steps of 2
const int q_offset = 32*im + l0;
const int s_offset = 8*im;
const int y_offset = 128*im + l0;
tmp[16 * ix + tid] = 0;
uint32_t aux[4];
const uint8_t * d = (const uint8_t *)aux;
const uint8_t * m = (const uint8_t *)(aux + 2);
for (int i = ix; i < num_blocks_per_row; i += K_QUANTS_PER_ITERATION) {
__global const float * y = yy + i * QK_K + y_offset;
// __global const uint8_t * q = BLOCK_Q2_K_QS(x, i, q_offset;
const float dall = BLOCK_Q2_K_D(x, i);
const float dmin = BLOCK_Q2_K_DMIN(x, i);
__global const uint32_t * a = (__global const uint32_t *)(x[i].scales + s_offset);
aux[0] = a[0] & 0x0f0f0f0f;
aux[1] = a[1] & 0x0f0f0f0f;
aux[2] = (a[0] >> 4) & 0x0f0f0f0f;
aux[3] = (a[1] >> 4) & 0x0f0f0f0f;
float sum1 = 0, sum2 = 0;
for (int l = 0; l < K_QUANTS_PER_ITERATION; ++l) {
sum1 += y[l+ 0] * d[0] * ((BLOCK_Q2_K_QS(x, i, q_offset + l+ 0) >> 0) & 3)
+ y[l+32] * d[2] * ((BLOCK_Q2_K_QS(x, i, q_offset + l+ 0) >> 2) & 3)
+ y[l+64] * d[4] * ((BLOCK_Q2_K_QS(x, i, q_offset + l+ 0) >> 4) & 3)
+ y[l+96] * d[6] * ((BLOCK_Q2_K_QS(x, i, q_offset + l+ 0) >> 6) & 3)
+ y[l+16] * d[1] * ((BLOCK_Q2_K_QS(x, i, q_offset + l+16) >> 0) & 3)
+ y[l+48] * d[3] * ((BLOCK_Q2_K_QS(x, i, q_offset + l+16) >> 2) & 3)
+ y[l+80] * d[5] * ((BLOCK_Q2_K_QS(x, i, q_offset + l+16) >> 4) & 3)
+y[l+112] * d[7] * ((BLOCK_Q2_K_QS(x, i, q_offset + l+16) >> 6) & 3);
sum2 += y[l+ 0] * m[0] + y[l+32] * m[2] + y[l+64] * m[4] + y[ l+96] * m[6]
+ y[l+16] * m[1] + y[l+48] * m[3] + y[l+80] * m[5] + y[l+112] * m[7];
}
tmp[16 * ix + tid] += dall * sum1 - dmin * sum2;
}
// sum up partial sums and write back result
barrier(CLK_LOCAL_MEM_FENCE);
for (int s=16; s>0; s>>=1) {
if (tid < s) {
tmp[tid] += tmp[tid + s];
}
barrier(CLK_LOCAL_MEM_FENCE);
}
if (tid == 0) {
dst[row] = tmp[0];
}
}
__kernel void dequantize_mul_mat_vec_q3_K(__global const struct block_q3_K * xx, __local float* tmp, __global float* yy, __global float* dst, const int ncols) {
const uint16_t kmask1 = 0x0303;
const uint16_t kmask2 = 0x0f0f;
const int row = get_group_id(0);
const int num_blocks_per_row = ncols / QK_K;
const int ib0 = row*num_blocks_per_row + get_global_offset(0);
__global const struct block_q3_K * x = xx + ib0;
const int tid = get_local_id(0)/K_QUANTS_PER_ITERATION; // 0...31 or 0...16
const int ix = get_local_id(0)%K_QUANTS_PER_ITERATION; // 0 or 0,1
const int n = K_QUANTS_PER_ITERATION; // iterations in the inner loop
const int step = 16/K_QUANTS_PER_ITERATION;
const int im = tid/step; // 0 or 1. 0 computes 0..., 1 computes 128...
const int in = tid - step*im; // 0....15 or 0...7
const uint8_t m = 1 << (4*im);
const int l0 = n*in; // 0...15 or 0...14 in steps of 2
const int q_offset = 32*im + l0;
const int y_offset = 128*im + l0;
uint16_t utmp[4];
const int8_t * s = (const int8_t *)utmp;
const uint16_t s_shift = 4*im;
tmp[16 * ix + tid] = 0;
for (int i = ix; i < num_blocks_per_row; i += K_QUANTS_PER_ITERATION) {
__global const float * y = yy + i * QK_K + y_offset;
__global const uint16_t * a = (__global const uint16_t *)x[i].scales;
utmp[0] = ((a[0] >> s_shift) & kmask2) | (((a[4] >> (s_shift + 0)) & kmask1) << 4);
utmp[1] = ((a[1] >> s_shift) & kmask2) | (((a[5] >> (s_shift + 0)) & kmask1) << 4);
utmp[2] = ((a[2] >> s_shift) & kmask2) | (((a[4] >> (s_shift + 2)) & kmask1) << 4);
utmp[3] = ((a[3] >> s_shift) & kmask2) | (((a[5] >> (s_shift + 2)) & kmask1) << 4);
const float d = BLOCK_Q2_K_D(x, i);
float sum = 0;
for (int l = 0; l < n; ++l) {
sum += y[l+ 0] * (s[0] - 32) * (((BLOCK_Q3_K_QS(x, i, q_offset + l) >> 0) & 3) - (BLOCK_Q3_K_HMASK(x, i, l0 + l) & (m << 0) ? 0 : 4))
+ y[l+32] * (s[2] - 32) * (((BLOCK_Q3_K_QS(x, i, q_offset + l) >> 2) & 3) - (BLOCK_Q3_K_HMASK(x, i, l0 + l) & (m << 1) ? 0 : 4))
+ y[l+64] * (s[4] - 32) * (((BLOCK_Q3_K_QS(x, i, q_offset + l) >> 4) & 3) - (BLOCK_Q3_K_HMASK(x, i, l0 + l) & (m << 2) ? 0 : 4))
+ y[l+96] * (s[6] - 32) * (((BLOCK_Q3_K_QS(x, i, q_offset + l) >> 6) & 3) - (BLOCK_Q3_K_HMASK(x, i, l0 + l) & (m << 3) ? 0 : 4));
sum += y[l+16] * (s[1] - 32) * (((BLOCK_Q3_K_QS(x, i, q_offset + l+16) >> 0) & 3) - (BLOCK_Q3_K_HMASK(x, i, l0 + l+16) & (m << 0) ? 0 : 4))
+ y[l+48] * (s[3] - 32) * (((BLOCK_Q3_K_QS(x, i, q_offset + l+16) >> 2) & 3) - (BLOCK_Q3_K_HMASK(x, i, l0 + l+16) & (m << 1) ? 0 : 4))
+ y[l+80] * (s[5] - 32) * (((BLOCK_Q3_K_QS(x, i, q_offset + l+16) >> 4) & 3) - (BLOCK_Q3_K_HMASK(x, i, l0 + l+16) & (m << 2) ? 0 : 4))
+ y[l+112] * (s[7] - 32) * (((BLOCK_Q3_K_QS(x, i, q_offset + l+16) >> 6) & 3) - (BLOCK_Q3_K_HMASK(x, i, l0 + l+16) & (m << 3) ? 0 : 4));
}
tmp[16 * ix + tid] += d * sum;
}
// sum up partial sums and write back result
barrier(CLK_LOCAL_MEM_FENCE);
for (int s=16; s>0; s>>=1) {
if (tid < s) {
tmp[tid] += tmp[tid + s];
}
barrier(CLK_LOCAL_MEM_FENCE);
}
if (tid == 0) {
dst[row] = tmp[0];
}
}
__kernel void dequantize_mul_mat_vec_q4_K(__global const struct block_q4_K * xx, __local float* tmp, __global float* yy, __global float* dst, const int ncols) {
//to rename it later, just to test now
const uint16_t kmask1 = 0x3f3f;
const uint16_t kmask2 = 0x0f0f;
const uint16_t kmask3 = 0xc0c0;
const int row = get_group_id(0);
const int num_blocks_per_row = ncols / QK_K;
const int ib0 = row*num_blocks_per_row + get_global_offset(0);
const int tid = get_local_id(0)/K_QUANTS_PER_ITERATION; // 0...15
const int ix = get_local_id(0)%K_QUANTS_PER_ITERATION;
const int step = 8/K_QUANTS_PER_ITERATION;
const int il = tid/step; // 0...3
const int ir = tid - step*il;// 0...3
const int n = 2*K_QUANTS_PER_ITERATION;
const int im = il/2; // 0 or 1. 0 computes 0,32 + 128,160, 1 computes 64,96 + 192,224
const int in = il%2;
const int l0 = n*(2*ir + in);
const int q_offset = 32*im + l0;
const int y_offset = 64*im + l0;
uint16_t aux[4];
const uint8_t * sc = (const uint8_t *)aux;
__global const struct block_q4_K * x = xx + ib0;
tmp[16 * ix + tid] = 0;
for (int i = ix; i < num_blocks_per_row; i += K_QUANTS_PER_ITERATION) {
__global const float * y1 = yy + i*QK_K + y_offset;
__global const float * y2 = y1 + 128;
const float dall = BLOCK_Q4_K_D(x, i);
const float dmin = BLOCK_Q4_K_DMIN(x, i);
__global const uint16_t * a = (__global const uint16_t *)x[i].scales;
aux[0] = a[im+0] & kmask1;
aux[1] = a[im+2] & kmask1;
aux[2] = ((a[im+4] >> 0) & kmask2) | ((a[im+0] & kmask3) >> 2);
aux[3] = ((a[im+4] >> 4) & kmask2) | ((a[im+2] & kmask3) >> 2);
float4 s = (float4)(0.f);
float smin = 0;
for (int l = 0; l < n; ++l) {
uint8_t q1 = BLOCK_Q4_K_QS(x, i, q_offset + l);
uint8_t q2 = BLOCK_Q4_K_QS(x, i, q_offset + 64 + l);
s.x += y1[l] * (q1 & 0xF); s.y += y1[l+32] * (q1 >> 4);
s.z += y2[l] * (q2 & 0xF); s.w += y2[l+32] * (q2 >> 4);
smin += y1[l] * sc[2] + y1[l+32] * sc[3] + y2[l] * sc[6] + y2[l+32] * sc[7];
}
tmp[16 * ix + tid] += dall * (s.x * sc[0] + s.y * sc[1] + s.z * sc[4] + s.w * sc[5]) - dmin * smin;
}
// sum up partial sums and write back result
barrier(CLK_LOCAL_MEM_FENCE);
for (int s=16; s>0; s>>=1) {
if (tid < s) {
tmp[tid] += tmp[tid + s];
}
barrier(CLK_LOCAL_MEM_FENCE);
}
if (tid == 0) {
dst[row] = tmp[0];
}
}
__kernel void dequantize_mul_mat_vec_q5_K(__global const struct block_q5_K * xx, __local float* tmp, __global float* yy, __global float* dst, const int ncols) {
const uint16_t kmask1 = 0x3f3f;
const uint16_t kmask2 = 0x0f0f;
const uint16_t kmask3 = 0xc0c0;
const int row = get_group_id(0);
const int num_blocks_per_row = ncols / QK_K;
const int ib0 = row*num_blocks_per_row + get_global_offset(0);
const int tid = get_local_id(0)/2; // 0...15
const int ix = get_local_id(0)%2;
const int il = tid/4; // 0...3
const int ir = tid - 4*il;// 0...3
const int n = 2;
const int im = il/2; // 0 or 1. 0 computes 0,32 + 128,160, 1 computes 64,96 + 192,224
const int in = il%2;
const int l0 = n*(2*ir + in);
const int q_offset = 32*im + l0;
const int y_offset = 64*im + l0;
const uint8_t hm1 = 1 << (2*im);
const uint8_t hm2 = hm1 << 4;
uint16_t aux[4];
const uint8_t * sc = (const uint8_t *)aux;
__global const struct block_q5_K * x = xx + ib0;
tmp[16 * ix + tid] = 0;
for (int i = ix; i < num_blocks_per_row; i += 2) {
__global const float * y1 = yy + i*QK_K + y_offset;
__global const float * y2 = y1 + 128;
const float dall = BLOCK_Q5_K_D(x, i);
const float dmin = BLOCK_Q5_K_DMIN(x, i);
__global const uint16_t * a = (__global const uint16_t *)x[i].scales;
aux[0] = a[im+0] & kmask1;
aux[1] = a[im+2] & kmask1;
aux[2] = ((a[im+4] >> 0) & kmask2) | ((a[im+0] & kmask3) >> 2);
aux[3] = ((a[im+4] >> 4) & kmask2) | ((a[im+2] & kmask3) >> 2);
float4 sum = (float4)(0.f);
float smin = 0;
for (int l = 0; l < n; ++l) {
sum.x += y1[l+ 0] * ((BLOCK_Q5_K_QS(x, i, q_offset + l+ 0) & 0xF) + (BLOCK_Q5_K_QH(x, i, l0 + l+ 0) & (hm1 << 0) ? 16 : 0))
+ y1[l+16] * ((BLOCK_Q5_K_QS(x, i, q_offset + l+16) & 0xF) + (BLOCK_Q5_K_QH(x, i, l0 + l+16) & (hm1 << 0) ? 16 : 0));
sum.y += y1[l+32] * ((BLOCK_Q5_K_QS(x, i, q_offset + l+ 0) >> 4) + (BLOCK_Q5_K_QH(x, i, l0 + l+ 0) & (hm1 << 1) ? 16 : 0))
+ y1[l+48] * ((BLOCK_Q5_K_QS(x, i, q_offset + l+16) >> 4) + (BLOCK_Q5_K_QH(x, i, l0 + l+16) & (hm1 << 1) ? 16 : 0));
sum.z += y2[l+ 0] * ((BLOCK_Q5_K_QS(x, i, q_offset + 64 + l+ 0) & 0xF) + (BLOCK_Q5_K_QH(x, i, l0 + l+ 0) & (hm2 << 0) ? 16 : 0))
+ y2[l+16] * ((BLOCK_Q5_K_QS(x, i, q_offset + 64 + l+16) & 0xF) + (BLOCK_Q5_K_QH(x, i, l0 + l+16) & (hm2 << 0) ? 16 : 0));
sum.w += y2[l+32] * ((BLOCK_Q5_K_QS(x, i, q_offset + 64 + l+ 0) >> 4) + (BLOCK_Q5_K_QH(x, i, l0 + l+ 0) & (hm2 << 1) ? 16 : 0))
+ y2[l+48] * ((BLOCK_Q5_K_QS(x, i, q_offset + 64 + l+16) >> 4) + (BLOCK_Q5_K_QH(x, i, l0 + l+16) & (hm2 << 1) ? 16 : 0));
smin += (y1[l] + y1[l+16]) * sc[2] + (y1[l+32] + y1[l+48]) * sc[3]
+ (y2[l] + y2[l+16]) * sc[6] + (y2[l+32] + y2[l+48]) * sc[7];
}
tmp[16 * ix + tid] += dall * (sum.x * sc[0] + sum.y * sc[1] + sum.z * sc[4] + sum.w * sc[5]) - dmin * smin;
}
// sum up partial sums and write back result
barrier(CLK_LOCAL_MEM_FENCE);
for (int s=16; s>0; s>>=1) {
if (tid < s) {
tmp[tid] += tmp[tid + s];
}
barrier(CLK_LOCAL_MEM_FENCE);
}
if (tid == 0) {
dst[row] = tmp[0];
}
}
__kernel void dequantize_mul_mat_vec_q6_K(__global const struct block_q6_K * x, __local float* tmp, __global const float * yy, __global float * dst, const int ncols) {
const int row = get_group_id(0);
const int num_blocks_per_row = ncols / QK_K;
const int ib0 = row*num_blocks_per_row + get_global_offset(0);
const int tid = get_local_id(0)/K_QUANTS_PER_ITERATION; // 0...31 or 0...16
const int ix = get_local_id(0)%K_QUANTS_PER_ITERATION; // 0 or 0, 1
const int step = 16/K_QUANTS_PER_ITERATION; // 16 or 8
const int im = tid/step; // 0 or 1. 0 computes 0..., 1 computes 128...
const int in = tid - step*im; // 0...15 or 0...7
#if K_QUANTS_PER_ITERATION == 1
const int l0 = K_QUANTS_PER_ITERATION*in; // 0...15
const int is = 0;
#else
const int l0 = 4 * in; // 0, 4, 8, ..., 28
const int is = in / 4;
#endif
const int ql_offset = 64*im + l0;
const int qh_offset = 32*im + l0;
const int s_offset = 8*im + is;
const int y_offset = 128*im + l0;
tmp[16 * ix + tid] = 0; // partial sum for thread in warp
for (int i = ix; i < num_blocks_per_row; i += K_QUANTS_PER_ITERATION) {
__global const float * y = yy + i * QK_K + y_offset;
int ii = ib0 + i;
const float d = BLOCK_Q6_K_D(x, ii);
// uint8_t ql = BLOCK_Q6_K_QL(x, ii, ql_offset + 0);
// uint8_t qh = BLOCK_Q6_K_QH(x, ii, qh_offset + 0);
// float sum = y[ 0] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 0) * d
// * ((int8_t)((ql & 0xF) | ((qh & 0x03) << 4)) - 32);
// tmp[16 * ix + tid] += sum;
#if K_QUANTS_PER_ITERATION == 1
float sum = y[ 0] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 0) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + 0) & 0xF) | ((BLOCK_Q6_K_QH(x, ii, qh_offset + 0) & 0x03) << 4)) - 32)
+ y[16] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 1) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + 16) & 0xF) | ((BLOCK_Q6_K_QH(x, ii, qh_offset + 16) & 0x03) << 4)) - 32)
+ y[32] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 2) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + 32) & 0xF) | ((BLOCK_Q6_K_QH(x, ii, qh_offset + 0) & 0x0c) << 2)) - 32)
+ y[48] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 3) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + 48) & 0xF) | ((BLOCK_Q6_K_QH(x, ii, qh_offset + 16) & 0x0c) << 2)) - 32)
+ y[64] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 4) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + 0) >> 4) | ((BLOCK_Q6_K_QH(x, ii, qh_offset + 0) & 0x30) >> 0)) - 32)
+ y[80] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 5) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + 16) >> 4) | ((BLOCK_Q6_K_QH(x, ii, qh_offset + 16) & 0x30) >> 0)) - 32)
+ y[96] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 6) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + 32) >> 4) | ((BLOCK_Q6_K_QH(x, ii, qh_offset + 0) & 0xc0) >> 2)) - 32)
+y[112] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 7) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + 48) >> 4) | ((BLOCK_Q6_K_QH(x, ii, qh_offset + 16) & 0xc0) >> 2)) - 32);
tmp[16 * ix + tid] += sum;
#else
float sum = 0;
for (int l = 0; l < 4; ++l) {
sum += y[l+ 0] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 0) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + l + 0) & 0xF) | (((BLOCK_Q6_K_QH(x, ii, qh_offset + l) >> 0) & 3) << 4)) - 32)
+ y[l+32] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 2) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + l + 32) & 0xF) | (((BLOCK_Q6_K_QH(x, ii, qh_offset + l) >> 2) & 3) << 4)) - 32)
+ y[l+64] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 4) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + l + 0) >> 4) | (((BLOCK_Q6_K_QH(x, ii, qh_offset + l) >> 4) & 3) << 4)) - 32)
+ y[l+96] * BLOCK_Q6_K_SCALES(x, ii, s_offset + 6) * d * ((int8_t)((BLOCK_Q6_K_QL(x, ii, ql_offset + l + 32) >> 4) | (((BLOCK_Q6_K_QH(x, ii, qh_offset + l) >> 6) & 3) << 4)) - 32);
}
tmp[16 * ix + tid] += sum;
#endif
}
// sum up partial sums and write back result
barrier(CLK_LOCAL_MEM_FENCE);
for (int s=16; s>0; s>>=1) {
if (tid < s) {
tmp[tid] += tmp[tid + s];
}
barrier(CLK_LOCAL_MEM_FENCE);
}
if (tid == 0) {
dst[row] = tmp[0];
}
}
#define DEQUANT(KERNEL_NAME, X_TYPE, QUANT_K, QUANT_R, DEQUANT_FUNC) \
__kernel void KERNEL_NAME(__global X_TYPE* x, __global float* y) { \
const size_t i = get_group_id(0)*get_local_size(0) + get_local_id(0)*2; \
\
if (i >= get_global_size(0)) { \
return; \
} \
\
const uint qk = QUANT_K; \
const uint qr = QUANT_R; \
\
const size_t ib = i/qk + get_global_offset(0); /* block index */ \
const size_t iqs = (i%qk)/qr; /* quant index */ \
const int iybs = i - i%qk; /* y block start index */ \
const int y_offset = qr == 1 ? 1 : qk/2; \
\
/* dequantize */ \
float v0, v1; \
DEQUANT_FUNC(x, ib, iqs, &v0, &v1); \
y[iybs + iqs + 0] = v0; \
y[iybs + iqs + y_offset] = v1; \
}
DEQUANT(dequantize_row_q4_0, struct block_q4_0, QK4_0, QR4_0, dequantize_q4_0)
DEQUANT(dequantize_row_q4_1, struct block_q4_1, QK4_1, QR4_1, dequantize_q4_1)
DEQUANT(dequantize_row_q5_0, struct block_q5_0, QK5_0, QR5_0, dequantize_q5_0)
DEQUANT(dequantize_row_q5_1, struct block_q5_1, QK5_1, QR5_1, dequantize_q5_1)
DEQUANT(dequantize_row_q8_0, struct block_q8_0, QK8_0, QR8_0, dequantize_q8_0)
#ifndef CLSPV_WORKAROUNDS
DEQUANT(convert_row_f16, half, 1, 1, convert_f16)
#else
DEQUANT(convert_row_f16, uint32_t, 1, 1, convert_f16)
#endif
#define DEQUANT_MUL_MAT_VEC(KERNEL_NAME, X_TYPE, QUANT_K, QUANT_R, DEQUANT_FUNC) \
__kernel void KERNEL_NAME(__global X_TYPE* x, __local float* tmp, __global float* y, __global float* dst, const int ncols) { \
const int local_size = get_local_size(0); \
const int row = get_group_id(0); \
const int tid = get_local_id(0); \
\
const uint qk = QUANT_K; \
const uint qr = QUANT_R; \
\
const int col_step = local_size * 2; \
const int y_offset = qr == 1 ? 1 : qk/2; \
\
x += get_global_offset(0); \
\
tmp[tid] = 0; \
\
for (int col = tid*2; col < ncols; col += col_step) { \
const int ib = (row*ncols + col)/qk; /* block index */ \
const int iqs = (col%qk)/qr; /* quant index */ \
const int iybs = col - col%qk; /* y block start index */ \
\
/* dequantize */ \
float v0, v1; \
DEQUANT_FUNC(x, ib, iqs, &v0, &v1); \
\
/* matrix multiplication */ \
tmp[tid] += v0 * y[iybs + iqs + 0]; \
tmp[tid] += v1 * y[iybs + iqs + y_offset]; \
} \
\
/* sum up partial sums and write back result */ \
barrier(CLK_LOCAL_MEM_FENCE); \
for (int s=local_size/2; s>0; s>>=1) { \
if (tid < s) { \
tmp[tid] += tmp[tid + s]; \
} \
barrier(CLK_LOCAL_MEM_FENCE); \
} \
if (tid == 0) { \
dst[row] = tmp[0]; \
} \
}
DEQUANT_MUL_MAT_VEC(dequantize_mul_mat_vec_q4_0, struct block_q4_0, QK4_0, QR4_0, dequantize_q4_0)
DEQUANT_MUL_MAT_VEC(dequantize_mul_mat_vec_q4_1, struct block_q4_1, QK4_1, QR4_1, dequantize_q4_1)
DEQUANT_MUL_MAT_VEC(dequantize_mul_mat_vec_q5_0, struct block_q5_0, QK5_0, QR5_0, dequantize_q5_0)
DEQUANT_MUL_MAT_VEC(dequantize_mul_mat_vec_q5_1, struct block_q5_1, QK5_1, QR5_1, dequantize_q5_1)
DEQUANT_MUL_MAT_VEC(dequantize_mul_mat_vec_q8_0, struct block_q8_0, QK8_0, QR8_0, dequantize_q8_0)
DEQUANT_MUL_MAT_VEC(convert_mul_mat_vec_f16, half, 1, 1, convert_f16);
#define MUL(KERNEL_NAME, TYPE) \
__kernel void KERNEL_NAME(__global TYPE* x, const int x_offset, __global TYPE* y, const int y_offset, __global TYPE* dst, const int dst_offset, const int ky) { \
const int i = get_group_id(0)*get_local_size(0) + get_local_id(0); \
\
if (i >= get_global_size(0)) { \
return; \
} \
\
dst[dst_offset + i] = x[x_offset + i] * y[y_offset + i%ky]; \
}
MUL(mul_f32, float)
Raw
ggml-opencl.cpp
#include "ggml-opencl.h"
#include <array>
#include <atomic>
#include <sstream>
#include <vector>
#include <limits>
#include <iostream>
#include <fstream>
#define CL_TARGET_OPENCL_VERSION 110
#include <clblast.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include "ggml.h"
#if defined(_MSC_VER)
#pragma warning(disable: 4244 4267) // possible loss of data
#endif
#define CL_DMMV_LOCAL_SIZE 32
#ifndef K_QUANTS_PER_ITERATION
#define K_QUANTS_PER_ITERATION 1
#else
static_assert(K_QUANTS_PER_ITERATION == 1 || K_QUANTS_PER_ITERATION == 2, "K_QUANTS_PER_ITERATION must be 1 or 2");
#endif
#define CL_CHECK(err) \
do { \
cl_int err_ = (err); \
if (err_ != CL_SUCCESS) { \
fprintf(stderr, "ggml_opencl: %s error %d at %s:%d\n", \
#err, err_, __FILE__, __LINE__); \
exit(1); \
} \
} while (0)
#define CLBLAST_CHECK(err) \
do { \
CLBlastStatusCode err_ = (err); \
if (err_ != CLBlastSuccess) { \
fprintf(stderr, "ggml_opencl: %s error %d at %s:%d\n", \
#err, err_, __FILE__, __LINE__); \
exit(1); \
} \
} while (0)
static std::string generate_kernels() {
std::string file = "ggml-opencl.cl";
// std::string file = "debug.cl";
std::ifstream kernels_file(file);
if (!kernels_file.is_open()) {
fprintf(stderr, "ggml_opencl: file %s not found!", file.c_str());
return "Kernels file not found.";
}
std::string line;
std::ostringstream src;
while (getline(kernels_file,line)) {
src << line << std::endl;
}
return src.str();
}
static cl_platform_id platform;
static cl_device_id device;
static cl_context context;
static cl_command_queue queue;
static cl_program program;
static cl_kernel convert_row_f16_cl;
static cl_kernel dequantize_row_q4_0_cl, dequantize_row_q4_1_cl, dequantize_row_q5_0_cl, dequantize_row_q5_1_cl, dequantize_row_q8_0_cl;
static cl_kernel dequantize_mul_mat_vec_q4_0_cl, dequantize_mul_mat_vec_q4_1_cl, dequantize_mul_mat_vec_q5_0_cl, dequantize_mul_mat_vec_q5_1_cl, dequantize_mul_mat_vec_q8_0_cl, convert_mul_mat_vec_f16_cl;
static cl_kernel dequantize_block_q2_k_cl, dequantize_block_q3_k_cl, dequantize_block_q4_k_cl, dequantize_block_q5_k_cl, dequantize_block_q6_k_cl;
static cl_kernel dequantize_mul_mat_vec_q2_K_cl, dequantize_mul_mat_vec_q3_K_cl, dequantize_mul_mat_vec_q4_K_cl, dequantize_mul_mat_vec_q5_K_cl, dequantize_mul_mat_vec_q6_K_cl;
static cl_kernel mul_f32_cl;
static bool fp16_support;
static bool byte_addressable_store_support;
static cl_program build_program_from_source(cl_context ctx, cl_device_id dev, const char* program_buffer) {
cl_program p;
char *program_log;
size_t program_size;
size_t log_size;
int err;
program_size = strlen(program_buffer);
p = clCreateProgramWithSource(ctx, 1, (const char**)&program_buffer, &program_size, &err);
if(err < 0) {
fprintf(stderr, "OpenCL error creating program");
exit(1);
}
std::string compile_opts = "-cl-mad-enable -cl-unsafe-math-optimizations -cl-finite-math-only -cl-fast-relaxed-math "
"-DQK4_0=32 -DQR4_0=2 -DQK4_1=32 -DQR4_1=2 -DQK5_0=32 -DQR5_0=2 -DQK5_1=32 -DQR5_1=2 -DQK8_0=32 -DQR8_0=1 "
"-DQK_K=256 -DK_QUANTS_PER_ITERATION=" + std::to_string(K_QUANTS_PER_ITERATION);
err = clBuildProgram(p, 0, NULL, compile_opts.c_str(), NULL, NULL);
if(err < 0) {
clGetProgramBuildInfo(p, dev, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size);
program_log = (char*) malloc(log_size + 1);
program_log[log_size] = '\0';
clGetProgramBuildInfo(p, dev, CL_PROGRAM_BUILD_LOG, log_size + 1, program_log, NULL);
fprintf(stderr, "ggml_opencl: kernel compile error:\n\n%s\n", program_log);
free(program_log);
exit(1);
}
return p;
}
void ggml_cl_init(void) {
cl_int err;
struct cl_device;
struct cl_platform {
cl_platform_id id;
unsigned number;
char name[128];
char vendor[128];
struct cl_device * devices;
unsigned n_devices;
struct cl_device * default_device;
};
struct cl_device {
struct cl_platform * platform;
cl_device_id id;
unsigned number;
cl_device_type type;
char name[128];
};
enum { NPLAT = 16, NDEV = 16 };
struct cl_platform platforms[NPLAT];
unsigned n_platforms = 0;
struct cl_device devices[NDEV];
unsigned n_devices = 0;
struct cl_device * default_device = NULL;
platform = NULL;
device = NULL;
cl_platform_id platform_ids[NPLAT];
CL_CHECK(clGetPlatformIDs(NPLAT, platform_ids, &n_platforms));
for (unsigned i = 0; i < n_platforms; i++) {
struct cl_platform * p = &platforms[i];
p->number = i;
p->id = platform_ids[i];
CL_CHECK(clGetPlatformInfo(p->id, CL_PLATFORM_NAME, sizeof(p->name), &p->name, NULL));
CL_CHECK(clGetPlatformInfo(p->id, CL_PLATFORM_VENDOR, sizeof(p->vendor), &p->vendor, NULL));
cl_device_id device_ids[NDEV];
cl_int clGetDeviceIDsError = clGetDeviceIDs(p->id, CL_DEVICE_TYPE_ALL, NDEV, device_ids, &p->n_devices);
if (clGetDeviceIDsError == CL_DEVICE_NOT_FOUND) {
p->n_devices = 0;
} else {
CL_CHECK(clGetDeviceIDsError);
}
p->devices = p->n_devices > 0 ? &devices[n_devices] : NULL;
p->default_device = NULL;
for (unsigned j = 0; j < p->n_devices; j++) {
struct cl_device * d = &devices[n_devices];
d->number = n_devices++;
d->id = device_ids[j];
d->platform = p;
CL_CHECK(clGetDeviceInfo(d->id, CL_DEVICE_NAME, sizeof(d->name), &d->name, NULL));
CL_CHECK(clGetDeviceInfo(d->id, CL_DEVICE_TYPE, sizeof(d->type), &d->type, NULL));
if (p->default_device == NULL && d->type == CL_DEVICE_TYPE_GPU) {
p->default_device = d;
}
}
if (default_device == NULL && p->default_device != NULL) {
default_device = p->default_device;
}
}
if (n_devices == 0) {
fprintf(stderr, "ggml_opencl: could find any OpenCL devices.\n");
exit(1);
}
char * user_platform_string = getenv("GGML_OPENCL_PLATFORM");
char * user_device_string = getenv("GGML_OPENCL_DEVICE");
int user_platform_number = -1;
int user_device_number = -1;
unsigned n;
if (user_platform_string != NULL && sscanf(user_platform_string, " %u", &n) == 1 && n < n_platforms) {
user_platform_number = (int)n;
}
if (user_device_string != NULL && sscanf(user_device_string, " %u", &n) == 1 && n < n_devices) {
user_device_number = (int)n;
}
if (user_platform_number != -1 && user_device_number != -1) {
cl_platform* platform = &platforms[user_platform_number];
if ((unsigned)user_device_number >= platform->n_devices) {
fprintf(stderr, "ggml_opencl: invalid device number %d\n", user_device_number);
exit(1);
}
default_device = &platform->devices[user_device_number];
} else {
struct cl_device * selected_devices = devices;
unsigned n_selected_devices = n_devices;
if (user_platform_number == -1 && user_platform_string != NULL && user_platform_string[0] != 0) {
for (unsigned i = 0; i < n_platforms; i++) {
struct cl_platform * p = &platforms[i];
if (strstr(p->name, user_platform_string) != NULL ||
strstr(p->vendor, user_platform_string) != NULL) {
user_platform_number = (int)i;
break;
}
}
if (user_platform_number == -1) {
fprintf(stderr, "ggml_opencl: no platform matching '%s' was found.\n", user_platform_string);
exit(1);
}
}
if (user_platform_number != -1) {
struct cl_platform * p = &platforms[user_platform_number];
selected_devices = p->devices;
n_selected_devices = p->n_devices;
default_device = p->default_device;
if (n_selected_devices == 0) {
fprintf(stderr, "ggml_opencl: selected platform '%s' does not have any devices.\n", p->name);
exit(1);
}
}
if (user_device_number == -1 && user_device_string != NULL && user_device_string[0] != 0) {
for (unsigned i = 0; i < n_selected_devices; i++) {
struct cl_device * d = &selected_devices[i];
if (strstr(d->name, user_device_string) != NULL) {
user_device_number = d->number;
break;
}
}
if (user_device_number == -1) {
fprintf(stderr, "ggml_opencl: no device matching '%s' was found.\n", user_device_string);
exit(1);
}
}
if (user_device_number != -1) {
selected_devices = &devices[user_device_number];
n_selected_devices = 1;
default_device = &selected_devices[0];
}
GGML_ASSERT(n_selected_devices > 0);
if (default_device == NULL) {
default_device = &selected_devices[0];
}
}
fprintf(stderr, "ggml_opencl: selecting platform: '%s'\n", default_device->platform->name);
fprintf(stderr, "ggml_opencl: selecting device: '%s'\n", default_device->name);
if (default_device->type != CL_DEVICE_TYPE_GPU) {
fprintf(stderr, "ggml_opencl: warning, not a GPU: '%s'.\n", default_device->name);
}
platform = default_device->platform->id;
device = default_device->id;
size_t ext_str_size;
clGetDeviceInfo(device, CL_DEVICE_EXTENSIONS, 0, NULL, &ext_str_size);
char *ext_buffer = (char *)alloca(ext_str_size + 1);
clGetDeviceInfo(device, CL_DEVICE_EXTENSIONS, ext_str_size, ext_buffer, NULL);
ext_buffer[ext_str_size] = '\0'; // ensure it is null terminated
// Check if ext_buffer contains cl_khr_fp16
fp16_support = strstr(ext_buffer, "cl_khr_fp16") != NULL;
byte_addressable_store_support = strstr(ext_buffer, "cl_khr_byte_addressable_store") != NULL;
fprintf(stderr, "ggml_opencl: device FP16 support: %s\n", fp16_support ? "true" : "false");
fprintf(stderr, "ggml_opencl: device Byte-addressable store support: %s\n", byte_addressable_store_support ? "true" : "false");
cl_context_properties properties[] = {
(intptr_t)CL_CONTEXT_PLATFORM, (intptr_t)platform, 0
};
CL_CHECK((context = clCreateContext(properties, 1, &device, NULL, NULL, &err), err));
CL_CHECK((queue = clCreateCommandQueue(context, device, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &err),
(err != CL_INVALID_QUEUE_PROPERTIES && err != CL_INVALID_VALUE ? err :
(queue = clCreateCommandQueue(context, device, 0, &err), err)
)));
const std::string kernel_src = generate_kernels();
program = build_program_from_source(context, device, kernel_src.c_str());
// FP16 to FP32 kernel
// CL_CHECK((convert_row_f16_cl = clCreateKernel(program, "convert_row_f16", &err), err));
convert_row_f16_cl = clCreateKernel(program, "convert_row_f16", &err); //
// Dequantize kernels
CL_CHECK((dequantize_row_q4_0_cl = clCreateKernel(program, "dequantize_row_q4_0", &err), err));
CL_CHECK((dequantize_row_q4_1_cl = clCreateKernel(program, "dequantize_row_q4_1", &err), err));
CL_CHECK((dequantize_row_q5_0_cl = clCreateKernel(program, "dequantize_row_q5_0", &err), err));
CL_CHECK((dequantize_row_q5_1_cl = clCreateKernel(program, "dequantize_row_q5_1", &err), err));
CL_CHECK((dequantize_row_q8_0_cl = clCreateKernel(program, "dequantize_row_q8_0", &err), err));
CL_CHECK((dequantize_row_q8_0_cl = clCreateKernel(program, "dequantize_row_q8_0", &err), err));
CL_CHECK((dequantize_block_q2_k_cl = clCreateKernel(program, "dequantize_block_q2_K", &err), err));
CL_CHECK((dequantize_block_q3_k_cl = clCreateKernel(program, "dequantize_block_q3_K", &err), err));
CL_CHECK((dequantize_block_q4_k_cl = clCreateKernel(program, "dequantize_block_q4_K", &err), err));
CL_CHECK((dequantize_block_q5_k_cl = clCreateKernel(program, "dequantize_block_q5_K", &err), err));
CL_CHECK((dequantize_block_q6_k_cl = clCreateKernel(program, "dequantize_block_q6_K", &err), err));
// dequant mul mat kernel
CL_CHECK((dequantize_mul_mat_vec_q4_0_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q4_0", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q4_1_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q4_1", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q5_0_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q5_0", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q5_1_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q5_1", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q8_0_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q8_0", &err), err));
CL_CHECK((convert_mul_mat_vec_f16_cl = clCreateKernel(program, "convert_mul_mat_vec_f16", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q2_K_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q2_K", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q3_K_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q3_K", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q4_K_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q4_K", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q5_K_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q5_K", &err), err));
CL_CHECK((dequantize_mul_mat_vec_q6_K_cl = clCreateKernel(program, "dequantize_mul_mat_vec_q6_K", &err), err));
// mul kernel
CL_CHECK((mul_f32_cl = clCreateKernel(program, "mul_f32", &err), err));
}
static cl_kernel* ggml_get_to_fp32_cl(ggml_type type) {
switch (type) {
case GGML_TYPE_Q4_0:
return &dequantize_row_q4_0_cl;
case GGML_TYPE_Q4_1:
return &dequantize_row_q4_1_cl;
case GGML_TYPE_Q5_0:
return &dequantize_row_q5_0_cl;
case GGML_TYPE_Q5_1:
return &dequantize_row_q5_1_cl;
case GGML_TYPE_Q8_0:
return &dequantize_row_q8_0_cl;
case GGML_TYPE_Q2_K:
return &dequantize_block_q2_k_cl;
case GGML_TYPE_Q3_K:
return &dequantize_block_q3_k_cl;
case GGML_TYPE_Q4_K:
return &dequantize_block_q4_k_cl;
case GGML_TYPE_Q5_K:
return &dequantize_block_q5_k_cl;
case GGML_TYPE_Q6_K:
return &dequantize_block_q6_k_cl;
case GGML_TYPE_F16:
return &convert_row_f16_cl;
default:
return nullptr;
}
}
static size_t ggml_cl_global_denom(ggml_type type) {
switch (type) {
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q5_0:
case GGML_TYPE_Q5_1:
case GGML_TYPE_Q8_0:
return 1;
case GGML_TYPE_Q2_K:
case GGML_TYPE_Q3_K:
return 4;
case GGML_TYPE_Q4_K:
return 8;
case GGML_TYPE_Q5_K:
case GGML_TYPE_Q6_K:
return 4;
case GGML_TYPE_F16:
default:
return 1;
}
}
static size_t ggml_cl_local_size(ggml_type type) {
switch (type) {
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q5_0:
case GGML_TYPE_Q5_1:
case GGML_TYPE_Q8_0:
return 0;
case GGML_TYPE_Q2_K:
case GGML_TYPE_Q3_K:
return 64;
case GGML_TYPE_Q4_K:
return 32;
case GGML_TYPE_Q5_K:
case GGML_TYPE_Q6_K:
return 64;
case GGML_TYPE_F16:
default:
return 0;
}
}
static cl_kernel* ggml_get_dequantize_mul_mat_vec_cl(ggml_type type) {
switch (type) {
case GGML_TYPE_Q4_0:
return &dequantize_mul_mat_vec_q4_0_cl;
case GGML_TYPE_Q4_1:
return &dequantize_mul_mat_vec_q4_1_cl;
case GGML_TYPE_Q5_0:
return &dequantize_mul_mat_vec_q5_0_cl;
case GGML_TYPE_Q5_1:
return &dequantize_mul_mat_vec_q5_1_cl;
case GGML_TYPE_Q8_0:
return &dequantize_mul_mat_vec_q8_0_cl;
case GGML_TYPE_F16:
return &convert_mul_mat_vec_f16_cl;
case GGML_TYPE_Q2_K:
return &dequantize_mul_mat_vec_q2_K_cl;
case GGML_TYPE_Q3_K:
return &dequantize_mul_mat_vec_q3_K_cl;
case GGML_TYPE_Q4_K:
return &dequantize_mul_mat_vec_q4_K_cl;
case GGML_TYPE_Q5_K:
return &dequantize_mul_mat_vec_q5_K_cl;
case GGML_TYPE_Q6_K:
return &dequantize_mul_mat_vec_q6_K_cl;
default:
return nullptr;
}
}
// buffer pool for cl
#define MAX_CL_BUFFERS 256
struct scoped_spin_lock {
std::atomic_flag& lock;
scoped_spin_lock(std::atomic_flag& lock) : lock(lock) {
while (lock.test_and_set(std::memory_order_acquire)) {
; // spin
}
}
~scoped_spin_lock() {
lock.clear(std::memory_order_release);
}
scoped_spin_lock(const scoped_spin_lock&) = delete;
scoped_spin_lock& operator=(const scoped_spin_lock&) = delete;
};
struct cl_buffer {
cl_mem mem;
size_t size = 0;
};
static cl_buffer g_cl_buffer_pool[MAX_CL_BUFFERS];
static std::atomic_flag g_cl_pool_lock = ATOMIC_FLAG_INIT;
static cl_mem ggml_cl_pool_malloc(size_t size, size_t * actual_size) {
scoped_spin_lock lock(g_cl_pool_lock);
cl_int err;
int best_i = -1;
size_t best_size = std::numeric_limits<size_t>::max(); //smallest unused buffer that fits our needs
int worst_i = -1;
size_t worst_size = 0; //largest unused buffer seen so far
for (int i = 0; i < MAX_CL_BUFFERS; ++i) {
cl_buffer &b = g_cl_buffer_pool[i];
if (b.size > 0 && b.size >= size && b.size < best_size)
{
best_i = i;
best_size = b.size;
}
if (b.size > 0 && b.size > worst_size)
{
worst_i = i;
worst_size = b.size;
}
}
if(best_i!=-1) //found the smallest buffer that fits our needs
{
cl_buffer& b = g_cl_buffer_pool[best_i];
cl_mem mem = b.mem;
*actual_size = b.size;
b.size = 0;
return mem;
}
if(worst_i!=-1) //no buffer that fits our needs, resize largest one to save memory
{
cl_buffer& b = g_cl_buffer_pool[worst_i];
cl_mem mem = b.mem;
b.size = 0;
clReleaseMemObject(mem);
}
cl_mem mem;
CL_CHECK((mem = clCreateBuffer(context, CL_MEM_READ_WRITE, size, NULL, &err), err));
*actual_size = size;
return mem;
}
static void ggml_cl_pool_free(cl_mem mem, size_t size) {
scoped_spin_lock lock(g_cl_pool_lock);
for (int i = 0; i < MAX_CL_BUFFERS; ++i) {
cl_buffer& b = g_cl_buffer_pool[i];
if (b.size == 0) {
b.mem = mem;
b.size = size;
return;
}
}
fprintf(stderr, "WARNING: cl buffer pool full, increase MAX_CL_BUFFERS\n");
clReleaseMemObject(mem);
}
void ggml_cl_free_data(const struct ggml_tensor* tensor) {
if (tensor->backend != GGML_BACKEND_GPU) {
return;
}
cl_mem mem = (cl_mem)tensor->extra;
clReleaseMemObject(mem);
}
static cl_int ggml_cl_h2d_tensor_2d(cl_command_queue queue, cl_mem dst, size_t offset, const struct ggml_tensor * src, uint64_t i3, uint64_t i2, cl_event* ev) {
cl_int err;
const uint64_t ne0 = src->ne[0];
const uint64_t ne1 = src->ne[1];
const uint64_t nb0 = src->nb[0];
const uint64_t nb1 = src->nb[1];
const uint64_t nb2 = src->nb[2];
const uint64_t nb3 = src->nb[3];
const enum ggml_type type = src->type;
const size_t ts = ggml_type_size(type);
const size_t bs = ggml_blck_size(type);
const uint64_t row_size = ts*ne0/bs;
const char * x = (const char *) src->data + i2*nb2 + i3*nb3;
if (nb0 == ts && nb1 == row_size) {
return clEnqueueWriteBuffer(queue, dst, CL_FALSE, offset, ne1*row_size, x, 0, NULL, ev);
}
if (nb0 == ts) {
const size_t buffer_origin[3] = { offset, 0, 0 };
const size_t host_origin[3] = { 0, 0, 0 };
const size_t region[3] = { row_size, ne1, 1 };
return clEnqueueWriteBufferRect(queue, dst, CL_FALSE, buffer_origin, host_origin, region, row_size, 0, nb1, 0, x, 0, NULL, ev);
}
std::vector<cl_event> events;
if (ev && ne1>1) events.reserve(ne1-1);
for (uint64_t i1 = 0; i1 < ne1; i1++) {
// pretend the row is a matrix with cols=1
const size_t buffer_origin[3] = { offset + i1*row_size, 0, 0 };
const size_t host_origin[3] = { 0, 0, 0 };
const size_t region[3] = { ts, ne0/bs, 1 };
// if an event is requested, make the last write wait for all previous writes to complete
if (ev && i1) {
events.push_back(*ev);
}
cl_uint nevents = i1 == ne1-1 ? events.size() : 0U;
err = clEnqueueWriteBufferRect(queue, dst, CL_FALSE, buffer_origin, host_origin, region, ts, 0, nb0, 0, x + i1*nb1, nevents, nevents ? events.data() : nullptr, ev);
if (err != CL_SUCCESS) {
for (auto event : events) {
clReleaseEvent(event);
}
return err;
}
}
for (auto event : events) {
CL_CHECK(clReleaseEvent(event));
}
return CL_SUCCESS;
}
static void ggml_cl_mul_f32(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
GGML_ASSERT(src1->backend == GGML_BACKEND_GPU);
const int64_t ne00 = src0->ne[0];
const int64_t ne01 = src0->ne[1];
const int64_t ne02 = src0->ne[2];
const int64_t ne03 = src0->ne[3];
const int64_t ne10 = src1->ne[0];
const int64_t ne11 = src1->ne[1];
const int64_t ne12 = src1->ne[2];
const int64_t ne13 = src1->ne[3];
const int nb2 = dst->nb[2];
const int nb3 = dst->nb[3];
size_t x_size;
size_t d_size;
cl_mem d_X = ggml_cl_pool_malloc(ne00 * ne01 * sizeof(float), &x_size); // src0
cl_mem d_Y = (cl_mem) src1->extra; // src1 is already on device, broadcasted.
cl_mem d_D = ggml_cl_pool_malloc(ne00 * ne01 * sizeof(float), &d_size); // dst
for (int64_t i03 = 0; i03 < ne03; i03++) {
for (int64_t i02 = 0; i02 < ne02; i02++) {
cl_event ev;
// copy src0 to device
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, d_X, 0, src0, i03, i02, &ev));
const int64_t i13 = i03%ne13;
const int64_t i12 = i02%ne12;
const int i1 = i13*ne12*ne11 + i12*ne11;
cl_int x_offset = 0;
cl_int y_offset = i1*ne10;
cl_int d_offset = 0;
size_t global = ne00 * ne01;
cl_int ky = ne10 * ne11;
CL_CHECK(clSetKernelArg(mul_f32_cl, 0, sizeof(cl_mem), &d_X));
CL_CHECK(clSetKernelArg(mul_f32_cl, 1, sizeof(cl_int), &x_offset));
CL_CHECK(clSetKernelArg(mul_f32_cl, 2, sizeof(cl_mem), &d_Y));
CL_CHECK(clSetKernelArg(mul_f32_cl, 3, sizeof(cl_int), &y_offset));
CL_CHECK(clSetKernelArg(mul_f32_cl, 4, sizeof(cl_mem), &d_D));
CL_CHECK(clSetKernelArg(mul_f32_cl, 5, sizeof(cl_int), &d_offset));
CL_CHECK(clSetKernelArg(mul_f32_cl, 6, sizeof(cl_int), &ky));
CL_CHECK(clEnqueueNDRangeKernel(queue, mul_f32_cl, 1, NULL, &global, NULL, 1, &ev, NULL));
CL_CHECK(clReleaseEvent(ev));
CL_CHECK(clFinish(queue));
// copy dst to host
float * d = (float *) ((char *) dst->data + i02*nb2 + i03*nb3);
CL_CHECK(clEnqueueReadBuffer(queue, d_D, true, 0, sizeof(float) * ne00*ne01, d, 0, NULL, NULL));
}
}
ggml_cl_pool_free(d_X, x_size);
ggml_cl_pool_free(d_D, d_size);
}
void ggml_cl_mul(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst) {
GGML_ASSERT(src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32);
ggml_cl_mul_f32(src0, src1, dst);
}
static void ggml_cl_mul_mat_f32(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
const int64_t ne00 = src0->ne[0];
const int64_t ne01 = src0->ne[1];
const int64_t ne02 = src0->ne[2];
const int64_t ne03 = src0->ne[3];
const int64_t ne10 = src1->ne[0];
const int64_t ne11 = src1->ne[1];
const int64_t ne12 = src1->ne[2];
const int64_t ne13 = src1->ne[3];
const int nb2 = dst->nb[2];
const int nb3 = dst->nb[3];
const int64_t r2 = ne12 / ne02;
const int64_t r3 = ne13 / ne03;
const float alpha = 1.0f;
const float beta = 0.0f;
const int x_ne = ne01 * ne00;
const int y_ne = ne11 * ne10;
const int d_ne = ne11 * ne01;
size_t x_size;
size_t y_size;
size_t d_size;
cl_mem d_X;
if (src0->backend == GGML_BACKEND_GPU) { // NOLINT
d_X = (cl_mem) src0->extra;
} else {
d_X = ggml_cl_pool_malloc(sizeof(float) * x_ne, &x_size);
}
cl_mem d_Y = ggml_cl_pool_malloc(sizeof(float) * y_ne, &y_size);
cl_mem d_D = ggml_cl_pool_malloc(sizeof(float) * d_ne, &d_size);
size_t x_offset = 0;
for (int64_t i03 = 0; i03 < ne03; i03++) {
// TODO: copy src0 here when r3>1
for (int64_t i13 = i03 * r3, e13 = i13 + r3; i13 < e13; i13++) {
for (int64_t i02 = 0; i02 < ne02; i02++) {
if (src0->backend == GGML_BACKEND_GPU) {
x_offset = (i03 * ne02 + i02) * x_ne;
} else {
// copy src0 to device
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, d_X, 0, src0, i03, i02, NULL));
}
for (int64_t i12 = i02 * r2, e12 = i12 + r2; i12 < e12; i12++) {
// copy src1 to device
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, d_Y, 0, src1, i13, i12, NULL));
CL_CHECK(clFinish(queue));
// compute
cl_event ev_sgemm;
clblast::StatusCode status = clblast::Gemm<cl_float>(clblast::Layout::kColMajor,
clblast::Transpose::kYes, clblast::Transpose::kNo,
ne01, ne11, ne10,
alpha,
d_X, x_offset, ne00,
d_Y, 0, ne10,
beta,
d_D, 0, ne01,
&queue, &ev_sgemm);
if (status != clblast::StatusCode::kSuccess) {
GGML_ASSERT(false);
}
// copy dst to host
float * d = (float *) ((char *) dst->data + i12*nb2 + i13*nb3);
CL_CHECK(clEnqueueReadBuffer(queue, d_D, true, 0, sizeof(float) * d_ne, d, 1, &ev_sgemm, NULL));
}
}
}
}
if (src0->backend != GGML_BACKEND_GPU) {
ggml_cl_pool_free(d_X, x_size);
}
ggml_cl_pool_free(d_Y, y_size);
ggml_cl_pool_free(d_D, d_size);
}
static void ggml_cl_mul_mat_f16(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, void * wdata, size_t wsize) {
GGML_ASSERT(fp16_support);
const int64_t ne00 = src0->ne[0];
const int64_t ne01 = src0->ne[1];
const int64_t ne02 = src0->ne[2];
const int64_t ne03 = src0->ne[3];
const int64_t ne10 = src1->ne[0];
const int64_t ne11 = src1->ne[1];
const int64_t ne12 = src1->ne[2];
const int64_t ne13 = src1->ne[3];
const int nb10 = src1->nb[0];
const int nb11 = src1->nb[1];
const int nb12 = src1->nb[2];
const int nb13 = src1->nb[3];
const int nb2 = dst->nb[2];
const int nb3 = dst->nb[3];
const int64_t r2 = ne12 / ne02;
const int64_t r3 = ne13 / ne03;
const ggml_fp16_t alpha = ggml_fp32_to_fp16(1.0f);
const ggml_fp16_t beta = ggml_fp32_to_fp16(0.0f);
const int x_ne = ne01 * ne00;
const int y_ne = ne11 * ne10;
const int d_ne = ne11 * ne01;
GGML_ASSERT(wsize >= sizeof(ggml_fp16_t) * y_ne);
GGML_ASSERT(wsize >= sizeof(ggml_fp16_t) * d_ne);
ggml_fp16_t * const tmp = (ggml_fp16_t *) wdata;
size_t x_size;
size_t y_size;
size_t d_size;
cl_mem d_X;
if (src0->backend == GGML_BACKEND_GPU) { // NOLINT
d_X = (cl_mem) src0->extra;
} else {
d_X = ggml_cl_pool_malloc(sizeof(ggml_fp16_t) * x_ne, &x_size);
}
cl_mem d_Y = ggml_cl_pool_malloc(sizeof(ggml_fp16_t) * y_ne, &y_size);
cl_mem d_D = ggml_cl_pool_malloc(sizeof(ggml_fp16_t) * d_ne, &d_size);
bool src1_cont_rows = nb10 == sizeof(float);
bool src1_cont_cols = (size_t)nb11 == ne11*sizeof(float);
size_t x_offset = 0;
for (int64_t i03 = 0; i03 < ne03; i03++) {
// TODO: copy src0 here when r3>1
for (int64_t i13 = i03 * r3, e13 = i13 + r3; i13 < e13; i13++) {
for (int64_t i02 = 0; i02 < ne02; i02++) {
if (src0->backend == GGML_BACKEND_GPU) {
x_offset = (i03 * ne02 + i02) * x_ne;
} else {
// copy src0 to device
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, d_X, 0, src0, i03, i02, NULL));
}
for (int64_t i12 = i02 * r2, e12 = i12 + r2; i12 < e12; i12++) {
// convert src1 to fp16
// TODO: use multiple threads
char * src1i = (char *) src1->data + i13*nb13 + i12*nb12;
if (src1_cont_rows) {
if (src1_cont_cols) {
ggml_fp32_to_fp16_row((float *) src1i, tmp, ne10*ne11);
}
else {
for (int64_t i11 = 0; i11 < ne11; i11++) {
ggml_fp32_to_fp16_row((float *) (src1i + i11*nb11), tmp + i11*ne10, ne10);
}
}
}
else {
for (int64_t i11 = 0; i11 < ne11; i11++) {
for (int64_t i10 = 0; i10 < ne10; i10++) {
// very slow due to no inlining
tmp[i11*ne10 + i10] = ggml_fp32_to_fp16(*(float *) (src1i + i11*nb11 + i10*nb10));
}
}
}
// copy src1 to device
CL_CHECK(clEnqueueWriteBuffer(queue, d_Y, false, 0, sizeof(ggml_fp16_t) * y_ne, tmp, 0, NULL, NULL));
CL_CHECK(clFinish(queue));
// compute
cl_event ev_sgemm;
clblast::StatusCode status = clblast::Gemm<cl_half>(clblast::Layout::kColMajor,
clblast::Transpose::kYes, clblast::Transpose::kNo,
ne01, ne11, ne10,
alpha,
d_X, x_offset, ne00,
d_Y, 0, ne10,
beta,
d_D, 0, ne01,
&queue, &ev_sgemm);
if (status != clblast::StatusCode::kSuccess) {
GGML_ASSERT(false);
}
// copy dst to host, then convert to float
CL_CHECK(clEnqueueReadBuffer(queue, d_D, true, 0, sizeof(ggml_fp16_t) * d_ne, tmp, 1, &ev_sgemm, NULL));
float * d = (float *) ((char *) dst->data + i12*nb2 + i13*nb3);
ggml_fp16_to_fp32_row(tmp, d, d_ne);
}
}
}
}
if (src0->backend != GGML_BACKEND_GPU) {
ggml_cl_pool_free(d_X, x_size);
}
ggml_cl_pool_free(d_Y, y_size);
ggml_cl_pool_free(d_D, d_size);
}
static void ggml_cl_mul_mat_q_f32(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
const int64_t ne00 = src0->ne[0];
const int64_t ne01 = src0->ne[1];
const int64_t ne02 = src0->ne[2];
const int64_t ne03 = src0->ne[3];
const int64_t ne10 = src1->ne[0];
const int64_t ne11 = src1->ne[1];
const int64_t ne12 = src1->ne[2];
const int64_t ne13 = src1->ne[3];
const int nb2 = dst->nb[2];
const int nb3 = dst->nb[3];
const ggml_type type = src0->type;
const bool mul_mat_vec = ne11 == 1 && ne00%2 == 0;
const int64_t r2 = ne12 / ne02;
const int64_t r3 = ne13 / ne03;
const float alpha = 1.0f;
const float beta = 0.0f;
const int x_ne = ne01 * ne00;
const int y_ne = ne11 * ne10;
const int d_ne = ne11 * ne01;
const int x_bps = x_ne / ggml_blck_size(type); // blocks per 2D slice
const size_t q_sz = ggml_type_size(type) * x_bps;
size_t x_size;
size_t y_size;
size_t d_size;
size_t q_size;
cl_mem d_X;
if (!mul_mat_vec) {
d_X = ggml_cl_pool_malloc(sizeof(float) * x_ne, &x_size);
}
cl_mem d_Y = ggml_cl_pool_malloc(sizeof(float) * y_ne, &y_size);
cl_mem d_D = ggml_cl_pool_malloc(sizeof(float) * d_ne, &d_size);
cl_mem d_Q;
if (src0->backend == GGML_BACKEND_CPU) {
d_Q = ggml_cl_pool_malloc(q_sz, &q_size);
}
cl_kernel* to_fp32_cl = ggml_get_to_fp32_cl(type);
cl_kernel* dmmv = ggml_get_dequantize_mul_mat_vec_cl(type);
GGML_ASSERT(to_fp32_cl != nullptr);
const size_t global_denom = ggml_cl_global_denom(type);
const size_t local = mul_mat_vec ? CL_DMMV_LOCAL_SIZE : ggml_cl_local_size(type);
size_t ev_idx = 0;
std::vector<cl_event> events;
for (int64_t i03 = 0; i03 < ne03; i03++) {
// TODO: copy and dequantize src0 here when r3>1
for (int64_t i13 = i03 * r3, e13 = i13 + r3; i13 < e13; i13++) {
for (int64_t i02 = 0; i02 < ne02; i02++) {
// copy src0 to device if necessary
if (src0->backend == GGML_BACKEND_CPU) {
events.emplace_back();
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, d_Q, 0, src0, i03, i02, events.data() + ev_idx++));
} else if (src0->backend == GGML_BACKEND_GPU) {
d_Q = (cl_mem) src0->extra;
} else {
GGML_ASSERT(false);
}
if (!mul_mat_vec) {
// convert src0 to fp32 on device
const size_t global = x_ne / global_denom;
const size_t offset = src0->backend == GGML_BACKEND_GPU ? (i03 * ne02 + i02) * x_bps : 0;
CL_CHECK(clSetKernelArg(*to_fp32_cl, 0, sizeof(cl_mem), &d_Q));
CL_CHECK(clSetKernelArg(*to_fp32_cl, 1, sizeof(cl_mem), &d_X));
CL_CHECK(clEnqueueNDRangeKernel(queue, *to_fp32_cl, 1, &offset, &global, local > 0 ? &local : NULL, events.size(), !events.empty() ? events.data() : NULL, NULL));
}
for (int64_t i12 = i02 * r2, e12 = i12 + r2; i12 < e12; i12++) {
if (mul_mat_vec) { // specialized dequantize_mul_mat_vec kernel
// copy src1 to device
events.emplace_back();
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, d_Y, 0, src1, i13, i12, events.data() + ev_idx++));
// compute
const size_t global = ne01 * local;
const size_t offset = src0->backend == GGML_BACKEND_GPU ? (i03 * ne02 + i02) * x_bps : 0;
const cl_int ncols = ne00;
events.emplace_back();
CL_CHECK(clSetKernelArg(*dmmv, 0, sizeof(cl_mem), &d_Q));
CL_CHECK(clSetKernelArg(*dmmv, 1, sizeof(float) * local, NULL));
CL_CHECK(clSetKernelArg(*dmmv, 2, sizeof(cl_mem), &d_Y));
CL_CHECK(clSetKernelArg(*dmmv, 3, sizeof(cl_mem), &d_D));
CL_CHECK(clSetKernelArg(*dmmv, 4, sizeof(cl_int), &ncols));
CL_CHECK(clEnqueueNDRangeKernel(queue, *dmmv, 1, &offset, &global, &local, events.size() - 1, events.data(), events.data() + ev_idx++));
} else { // CLBlast matrix matrix multiplication
// copy src1 to device
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, d_Y, 0, src1, i13, i12, NULL));
// wait for conversion
CL_CHECK(clFinish(queue));
// compute
events.emplace_back();
clblast::StatusCode status = clblast::Gemm<cl_float>(clblast::Layout::kColMajor,
clblast::Transpose::kYes, clblast::Transpose::kNo,
ne01, ne11, ne10,
alpha,
d_X, 0, ne00,
d_Y, 0, ne10,
beta,
d_D, 0, ne01,
&queue, events.data() + ev_idx++);
if (status != clblast::StatusCode::kSuccess) {
GGML_ASSERT(false);
}
}
// copy dst to host
float * d = (float *) ((char *) dst->data + i12*nb2 + i13*nb3);
CL_CHECK(clEnqueueReadBuffer(queue, d_D, true, 0, sizeof(float) * d_ne, d, 1, &events[events.size() - 1], NULL));
for (auto *event : events) {
clReleaseEvent(event);
}
ev_idx = 0;
events.clear();
}
}
}
}
if (!mul_mat_vec) {
ggml_cl_pool_free(d_X, x_size);
}
ggml_cl_pool_free(d_Y, y_size);
ggml_cl_pool_free(d_D, d_size);
if (src0->backend == GGML_BACKEND_CPU) {
ggml_cl_pool_free(d_Q, q_size);
}
}
bool ggml_cl_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst) {
const int64_t ne10 = src1->ne[0];
const int64_t ne0 = dst->ne[0];
const int64_t ne1 = dst->ne[1];
// TODO: find the optimal values for these
if ((src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16 || ggml_is_quantized(src0->type)) &&
src1->type == GGML_TYPE_F32 &&
dst->type == GGML_TYPE_F32 &&
((ne0 >= 32 && ne1 >= 32 && ne10 >= 32) || src0->backend == GGML_BACKEND_GPU)) {
return true;
}
return false;
}
static bool ggml_cl_mul_mat_use_f16(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * /* dst */) {
// If device doesn't support FP16
if (!fp16_support) {
return false;
}
size_t src0_sz = ggml_nbytes(src0);
size_t src1_sz = ggml_nbytes(src1);
// mul_mat_q: src0 is converted to fp32 on device
size_t mul_mat_q_transfer = src0_sz + src1_sz;
// mul_mat_f16: src1 is converted to fp16 on cpu
size_t mul_mat_f16_transfer = src0_sz + sizeof(ggml_fp16_t) * ggml_nelements(src1);
// choose the smaller one to transfer to the device
// TODO: this is not always the best choice due to the overhead of converting to fp16
return mul_mat_f16_transfer < mul_mat_q_transfer;
}
void ggml_cl_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst, void * wdata, size_t wsize) {
GGML_ASSERT(ggml_cl_can_mul_mat(src0, src1, dst));
if (src0->type == GGML_TYPE_F32) {
ggml_cl_mul_mat_f32(src0, src1, dst);
}
else if (src0->type == GGML_TYPE_F16) {
if (ggml_cl_mul_mat_use_f16(src0, src1, dst)) {
ggml_cl_mul_mat_f16(src0, src1, dst, wdata, wsize);
}
else {
ggml_cl_mul_mat_q_f32(src0, src1, dst);
}
}
else if (ggml_is_quantized(src0->type)) {
ggml_cl_mul_mat_q_f32(src0, src1, dst);
}
else {
GGML_ASSERT(false);
}
}
size_t ggml_cl_mul_mat_get_wsize(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst) {
if (src0->type == GGML_TYPE_F16 && ggml_cl_mul_mat_use_f16(src0, src1, dst)) {
return sizeof(ggml_fp16_t) * std::max(src1->ne[0] * src1->ne[1], dst->ne[0] * dst->ne[1]);
}
return 0;
}
void ggml_cl_transform_tensor(void * data, ggml_tensor * tensor) {
const int64_t ne0 = tensor->ne[0];
const int64_t ne1 = tensor->ne[1];
const int64_t ne2 = tensor->ne[2];
const int64_t ne3 = tensor->ne[3];
const ggml_type type = tensor->type;
const size_t s_sz = ggml_type_size(type) * (size_t) (ne0 * ne1 / ggml_blck_size(type));
const size_t q_sz = s_sz * (size_t) (ne2 * ne3);
size_t q_size;
cl_mem dst = ggml_cl_pool_malloc(q_sz, &q_size);
tensor->data = data;
// copy tensor to device
size_t offset = 0;
for (int64_t i3 = 0; i3 < ne3; i3++) {
for (int64_t i2 = 0; i2 < ne2; i2++) {
CL_CHECK(ggml_cl_h2d_tensor_2d(queue, dst, offset, tensor, i3, i2, NULL));
offset += s_sz;
}
}
CL_CHECK(clFinish(queue));
tensor->extra = dst;
GGML_ASSERT(tensor->backend == GGML_BACKEND_GPU);
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment