nicolas
/
ik_llama_opt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Q8_K_R8: Fastest quantized matrix multiplications (#141) 

* q8_k_r8: fastest matrix multiplication known to human kind

We get PP-512(LLaMA-3.1-8B) = 370 t/s on a Ryzen-7950X!

* q8_k_r8: AVX2

I was worried that we don't have enough vector registrers on
AVX2, but it looks like it handles it just fine. We get
PP-512(LLaMA-3.1-8B) = 354 t/s on a Ryzen-5975WX.
Slightly slower than the Zen4 version with double the threads,
but still a huge upgrade compared to Q8_0_R4.

* q8_k_r4: NEON

We get PP-512(LLaMA-3.1-8B) = 159.2 t/s.
Compare this to the 128 t/s we have fr Q8_0_R4.

* q8_k_r4: go to signed ints

Why?
* On AVX2 _mm256_maddubs_epi16() may overflow, so we need to
  stay within the signed int range and use _mm256_sign_epi8.
  Not yet tested on the AVX2 comp, vut expect major slowdown.
* It is almost 10% faster on ARM_NEON. Somehow the veorrq_u8()
  needed tto convert from unsigned to signed seems to be extremely
  slow on the M2-Max
* We only lose ~0.5% in oerformance on Zen4 (there the exclusive
  or that we now use to convert fro signed to unsigned seems to be
  much faster than on M2-Max)

* Shutup useless compiler warnings

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow 2024-12-14 09:24:30 +01:00 committed by GitHub
parent eae584dc98
commit e885c1e59b
10 changed files with 301 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
examples/quantize/quantize.cpp
View File

@ -70,6 +70,7 @@ static const std::vector<struct quant_option> QUANT_OPTIONS = {
{ "Q5_K_M", LLAMA_FTYPE_MOSTLY_Q5_K_M, " 4.45G, +0.0122 ppl @ LLaMA-v1-7B", },
{ "Q6_K", LLAMA_FTYPE_MOSTLY_Q6_K, " 5.15G, +0.0008 ppl @ LLaMA-v1-7B", },
{ "Q6_K_R4", LLAMA_FTYPE_MOSTLY_Q6_K_R4, "Q6_K repacked", },
{ "Q8_K_R8", LLAMA_FTYPE_MOSTLY_Q8_K_R8, "Q8_K repacked", },
{ "Q8_0", LLAMA_FTYPE_MOSTLY_Q8_0, " 6.70G, +0.0004 ppl @ LLaMA-v1-7B", },
{ "Q4_0_4_4", LLAMA_FTYPE_MOSTLY_Q4_0_4_4, " 4.34G, +0.4685 ppl @ Llama-3-8B", },
{ "Q4_0_4_8", LLAMA_FTYPE_MOSTLY_Q4_0_4_8, " 4.34G, +0.4685 ppl @ Llama-3-8B", },

3
ggml/include/ggml.h
View File

@ -408,6 +408,7 @@ extern "C" {
GGML_TYPE_IQ4_KSS = 146,
GGML_TYPE_Q8_K16 = 147,
GGML_TYPE_Q8_K32 = 148,
GGML_TYPE_Q8_KR8 = 149,
GGML_TYPE_Q4_0_R4 = 202,
GGML_TYPE_Q5_0_R4 = 206,
@ -422,6 +423,7 @@ extern "C" {
GGML_TYPE_Q6_0_R4 = 233,
GGML_TYPE_IQ2_BN_R4 = 335,
GGML_TYPE_IQ4_K_R4 = 339,
GGML_TYPE_Q8_K_R8 = 399,
GGML_TYPE_COUNT,
};
@ -494,6 +496,7 @@ extern "C" {
GGML_FTYPE_MOSTLY_Q6_0_R4 = 227, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ2_BN_R4 = 329, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ4_K_R4 = 332, // except 1d tensors
GGML_FTYPE_MOSTLY_Q8_K_R8 = 399, // except 1d tensors
};
// available tensor operations:

6
ggml/src/ggml-common.h
View File

@ -382,6 +382,12 @@ typedef struct {
} block_q8_K128;
static_assert(sizeof(block_q8_K128) == sizeof(float) + 128, "wrong q8_K128 block size/padding");
typedef struct {
ggml_half d[8]; // delta
int8_t qs[8*QK_K]; // quants, stored as unsigned ints
} block_q8_k_r8;
static_assert(sizeof(block_q8_k_r8) == 8*sizeof(ggml_half) + 8*QK_K, "wrong q8_k_r8 block size/padding");
// (Almost) "true" 2-bit quantization.
// Due to the need to use blocks as per ggml design, it ends up using
// 2.0625 bpw because of the 16-bit scale for each block of 256.

1
ggml/src/ggml-quants.c
View File

@ -15208,6 +15208,7 @@ bool ggml_validate_row_data(enum ggml_type type, const void * data, size_t nbyte
case GGML_TYPE_Q5_K_R4: break;
case GGML_TYPE_Q6_K_R4: break;
case GGML_TYPE_IQ4_K_R4: break;
case GGML_TYPE_Q8_K_R8: break;
case GGML_TYPE_Q4_0_4_4:
case GGML_TYPE_Q4_0_4_8:
{

31
ggml/src/ggml.c
View File

@ -979,6 +979,19 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
.nrows = 1,
.row_meta_size = 0,
},
[GGML_TYPE_Q8_K_R8] = {
.type_name = "q8_k_r8",
.blck_size = QK_K,
.type_size = sizeof(block_q8_k_r8)/8,
.is_quantized = true,
.to_float = (ggml_to_float_t) dequantize_row_q8_k_r8,
.from_float = quantize_row_q8_k_r8,
.from_float_ref = (ggml_from_float_t) quantize_row_q8_k_r8_ref,
.vec_dot = vec_dot_q8_k_r8_q8_k,
.vec_dot_type = GGML_TYPE_Q8_KR8,
.nrows = 1,
.row_meta_size = 0,
},
[GGML_TYPE_IQ2_XXS] = {
.type_name = "iq2_xxs",
.blck_size = QK_K,
@ -1197,6 +1210,14 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
.from_float = quantize_row_q8_K32,
.row_meta_size = 0,
},
[GGML_TYPE_Q8_KR8] = {
.type_name = "q8_KR8",
.blck_size = QK_K,
.type_size = sizeof(block_q8_K),
.is_quantized = true,
.from_float = quantize_row_q8_KR8,
.row_meta_size = 0,
},
[GGML_TYPE_BF16] = {
.type_name = "bf16",
.blck_size = 1,
@ -4105,6 +4126,7 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) {
case GGML_FTYPE_MOSTLY_Q5_K_R4: wtype = GGML_TYPE_Q5_K_R4; break;
case GGML_FTYPE_MOSTLY_Q6_K: wtype = GGML_TYPE_Q6_K; break;
case GGML_FTYPE_MOSTLY_Q6_K_R4: wtype = GGML_TYPE_Q6_K_R4; break;
case GGML_FTYPE_MOSTLY_Q8_K_R8: wtype = GGML_TYPE_Q8_K_R8; break;
case GGML_FTYPE_MOSTLY_IQ2_XXS: wtype = GGML_TYPE_IQ2_XXS; break;
case GGML_FTYPE_MOSTLY_IQ2_XS: wtype = GGML_TYPE_IQ2_XS; break;
case GGML_FTYPE_MOSTLY_IQ3_XXS: wtype = GGML_TYPE_IQ3_XXS; break;
@ -10641,6 +10663,7 @@ static void ggml_compute_forward_add(
case GGML_TYPE_Q5_K_R4:
case GGML_TYPE_Q6_K:
case GGML_TYPE_Q6_K_R4:
case GGML_TYPE_Q8_K_R8:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ3_XXS:
@ -11096,6 +11119,7 @@ static void ggml_compute_forward_add1(
case GGML_TYPE_Q5_K_R4:
case GGML_TYPE_Q6_K:
case GGML_TYPE_Q6_K_R4:
case GGML_TYPE_Q8_K_R8:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ3_XXS:
@ -11248,6 +11272,7 @@ static void ggml_compute_forward_acc(
case GGML_TYPE_Q5_K_R4:
case GGML_TYPE_Q6_K:
case GGML_TYPE_Q6_K_R4:
case GGML_TYPE_Q8_K_R8:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ3_XXS:
@ -14446,6 +14471,7 @@ static void ggml_compute_forward_out_prod(
case GGML_TYPE_Q5_K_R4:
case GGML_TYPE_Q6_K:
case GGML_TYPE_Q6_K_R4:
case GGML_TYPE_Q8_K_R8:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ3_XXS:
@ -14838,6 +14864,7 @@ static void ggml_compute_forward_set(
case GGML_TYPE_Q5_K_R4:
case GGML_TYPE_Q6_K:
case GGML_TYPE_Q6_K_R4:
case GGML_TYPE_Q8_K_R8:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ3_XXS:
@ -15124,6 +15151,7 @@ static void ggml_compute_forward_get_rows(
case GGML_TYPE_Q5_K_R4:
case GGML_TYPE_Q6_K:
case GGML_TYPE_Q6_K_R4:
case GGML_TYPE_Q8_K_R8:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ3_XXS:
@ -15737,6 +15765,8 @@ static void ggml_compute_forward_clamp(
case GGML_TYPE_Q5_K_R4:
case GGML_TYPE_Q6_K:
case GGML_TYPE_Q6_K_R4:
case GGML_TYPE_Q8_K_R8:
case GGML_TYPE_Q8_KR8:
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_IQ3_XXS:
@ -22578,6 +22608,7 @@ size_t ggml_quantize_chunk(
case GGML_TYPE_Q5_K_R4: result = quantize_q5_k_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_Q6_K: result = quantize_q6_K(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_Q6_K_R4: result = quantize_q6_k_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_Q8_K_R8: result = quantize_q8_k_r8(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ2_XXS: result = quantize_iq2_xxs(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ2_XS: result = quantize_iq2_xs (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ3_XXS: result = quantize_iq3_xxs(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;

139
ggml/src/iqk/iqk_mul_mat.cpp
View File

@ -92,6 +92,9 @@ struct DataInfo {
inline void store(int ix, int iy, __m128 result) const {
_mm_storeu_ps(dst_row(iy) + ix, result);
}
inline void store(int ix, int iy, __m256 result) const {
_mm256_storeu_ps(dst_row(iy) + ix, result);
}
#endif
#ifdef __ARM_NEON
inline void store(int ix, int iy, float32x4_t result) const {
@ -175,6 +178,7 @@ struct MulMat {
case GGML_TYPE_IQ4_NL_R4:
case GGML_TYPE_IQ4_XS_R4:
case GGML_TYPE_IQ2_BN_R4: return 4;
case GGML_TYPE_Q8_K_R8: return 8;
default: return 1;
}
}
@ -3802,6 +3806,76 @@ static void mul_mat_q6_k_r4_q8_k(int n, const void * vx, size_t bx, const DataIn
}
}
// The HAVE_FANCY_SIMD should only be #if defined(__AVX512_VNNI__ && defined(__AVX512VL__)
template <int nrc_y>
static void mul_mat_q8_k_r8_q8_k(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) {
GGML_ASSERT(nrc_x%4 == 0);
Q8<nrc_y, block_q8_K> q8(info);
#ifndef HAVE_FANCY_SIMD
auto m1 = _mm256_set1_epi16(1);
#endif
int nbl = n / QK_K;
__m256 acc[nrc_y] = {};
__m256i isum[nrc_y] = {};
__m256i qx[4];
for (int ix = 0; ix < nrc_x; ix += 8) {
const block_q8_k_r8 * iq8 = (const block_q8_k_r8 *)((const char *)vx + (ix+0)*bx);
for (int ibl = 0; ibl < nbl; ++ibl) { // Block of 256
auto d4 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i *)iq8[ibl].d));
for (int ib = 0; ib < QK_K/16; ++ib) {
qx[0] = _mm256_loadu_si256((const __m256i *)iq8[ibl].qs+4*ib+0);
qx[1] = _mm256_loadu_si256((const __m256i *)iq8[ibl].qs+4*ib+1);
qx[2] = _mm256_loadu_si256((const __m256i *)iq8[ibl].qs+4*ib+2);
qx[3] = _mm256_loadu_si256((const __m256i *)iq8[ibl].qs+4*ib+3);
#ifdef HAVE_FANCY_SIMD
qx[0] = _mm256_xor_si256(_mm256_loadu_si256((const __m256i *)iq8[ibl].qs+4*ib+0), _mm256_set1_epi8(-128));
qx[1] = _mm256_xor_si256(_mm256_loadu_si256((const __m256i *)iq8[ibl].qs+4*ib+1), _mm256_set1_epi8(-128));
qx[2] = _mm256_xor_si256(_mm256_loadu_si256((const __m256i *)iq8[ibl].qs+4*ib+2), _mm256_set1_epi8(-128));
qx[3] = _mm256_xor_si256(_mm256_loadu_si256((const __m256i *)iq8[ibl].qs+4*ib+3), _mm256_set1_epi8(-128));
#else
auto s0 = _mm256_sign_epi8(qx[0], qx[0]);
auto s1 = _mm256_sign_epi8(qx[1], qx[1]);
auto s2 = _mm256_sign_epi8(qx[2], qx[2]);
auto s3 = _mm256_sign_epi8(qx[3], qx[3]);
#endif
for (int iy = 0; iy < nrc_y; ++iy) {
auto y128 = _mm_loadu_si128((const __m128i*)q8.y[iy][ibl].qs+ib);
auto y = MM256_SET_M128I(y128, y128);
#ifdef HAVE_FANCY_SIMD
isum[iy] = _mm256_dpbusd_epi32(isum[iy], qx[0], _mm256_shuffle_epi32(y, 0x00));
isum[iy] = _mm256_dpbusd_epi32(isum[iy], qx[1], _mm256_shuffle_epi32(y, 0x55));
isum[iy] = _mm256_dpbusd_epi32(isum[iy], qx[2], _mm256_shuffle_epi32(y, 0xaa));
isum[iy] = _mm256_dpbusd_epi32(isum[iy], qx[3], _mm256_shuffle_epi32(y, 0xff));
#else
auto sumi1 = _mm256_madd_epi16(m1, _mm256_maddubs_epi16(s0, _mm256_sign_epi8(_mm256_shuffle_epi32(y, 0x00), qx[0])));
auto sumi2 = _mm256_madd_epi16(m1, _mm256_maddubs_epi16(s1, _mm256_sign_epi8(_mm256_shuffle_epi32(y, 0x55), qx[1])));
auto sumi3 = _mm256_madd_epi16(m1, _mm256_maddubs_epi16(s2, _mm256_sign_epi8(_mm256_shuffle_epi32(y, 0xaa), qx[2])));
auto sumi4 = _mm256_madd_epi16(m1, _mm256_maddubs_epi16(s3, _mm256_sign_epi8(_mm256_shuffle_epi32(y, 0xff), qx[3])));
isum[iy] = _mm256_add_epi32(isum[iy], _mm256_add_epi32(sumi1, sumi2));
isum[iy] = _mm256_add_epi32(isum[iy], _mm256_add_epi32(sumi3, sumi4));
#endif
}
}
#ifdef HAVE_FANCY_SIMD
auto m4 = _mm256_mul_ps(d4, _mm256_set1_ps(-128.f));
#endif
for (int iy = 0; iy < nrc_y; ++iy) {
auto d4y = _mm256_mul_ps(d4, _mm256_set1_ps(q8.scale(iy, ibl)));
acc[iy] = _mm256_fmadd_ps(d4y, _mm256_cvtepi32_ps(isum[iy]), acc[iy]);
#ifdef HAVE_FANCY_SIMD
auto bsums = (const float *)q8.y[iy][ibl].bsums;
acc[iy] = _mm256_fmadd_ps(m4, _mm256_set1_ps(bsums[0]), acc[iy]);
#endif
isum[iy] = _mm256_setzero_si256();
}
}
for (int iy = 0; iy < nrc_y; ++iy) {
info.store(ix, iy, acc[iy]);
acc[iy] = _mm256_setzero_ps();
}
}
}
template <int nrc_y>
static void mul_mat_iq4_k_r4_q8_k(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) {
GGML_ASSERT(nrc_x%4 == 0);
@ -5976,6 +6050,18 @@ bool MulMat::prepare(int typeA, int typeB, int ne00, MulMat& mm, int Ny) {
mm.funcs[7] = mul_mat_q6_k_r4_q8_k<8>;
expected_typeB = GGML_TYPE_Q8_K;
break;
case GGML_TYPE_Q8_K_R8:
assert (ne00 % QK_K == 0);
mm.funcs[0] = mul_mat_q8_k_r8_q8_k<1>;
mm.funcs[1] = mul_mat_q8_k_r8_q8_k<2>;
mm.funcs[2] = mul_mat_q8_k_r8_q8_k<3>;
mm.funcs[3] = mul_mat_q8_k_r8_q8_k<4>;
mm.funcs[4] = mul_mat_q8_k_r8_q8_k<5>;
mm.funcs[5] = mul_mat_q8_k_r8_q8_k<6>;
mm.funcs[6] = mul_mat_q8_k_r8_q8_k<7>;
mm.funcs[7] = mul_mat_q8_k_r8_q8_k<8>;
expected_typeB = GGML_TYPE_Q8_KR8;
break;
case GGML_TYPE_IQ4_K_R4:
assert (ne00 % QK_K == 0);
mm.funcs[0] = mul_mat_iq4_k_r4_q8_k<1>;
@ -9158,6 +9244,55 @@ void mul_mat_q6_k_r4_q8_k(int n, const void * vx, size_t bx, const DataInfo& inf
}
}
template <int nrc_y>
void mul_mat_q8_k_r8_q8_k(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) {
GGML_ASSERT(nrc_x%8 == 0);
Q8<nrc_y, block_q8_K> q8(info);
int nbl = n / QK_K;
float32x4_t acc[2*nrc_y] = {};
for (int ix = 0; ix < nrc_x; ix += 8) {
const block_q8_k_r8 * iq8 = (const block_q8_k_r8 *)((const char *)vx + ix*bx);
for (int ibl = 0; ibl < nbl; ++ibl) {
auto d4l = vcvt_f32_f16(vld1_f16((const float16_t *)iq8[ibl].d+0));
auto d4h = vcvt_f32_f16(vld1_f16((const float16_t *)iq8[ibl].d+4));
int32x4_t isum[2*nrc_y] = {};
for (int ib = 0; ib < QK_K/16; ++ib) {
auto q1 = vld1q_u8_x4(iq8[ibl].qs + 128*ib + 0);
auto q2 = vld1q_u8_x4(iq8[ibl].qs + 128*ib + 64);
for (int k = 0; k < 4; ++k) {
q1.val[k] = veorq_u8(q1.val[k], vdupq_n_u8(0x80));
q2.val[k] = veorq_u8(q2.val[k], vdupq_n_u8(0x80));
}
for (int iy = 0; iy < nrc_y; ++iy) {
auto y = vld1q_s8(q8.y[iy][ibl].qs+16*ib);
isum[2*iy+0] = vdotq_laneq_s32(isum[2*iy+0], q1.val[0], y, 0);
isum[2*iy+1] = vdotq_laneq_s32(isum[2*iy+1], q1.val[1], y, 0);
isum[2*iy+0] = vdotq_laneq_s32(isum[2*iy+0], q1.val[2], y, 1);
isum[2*iy+1] = vdotq_laneq_s32(isum[2*iy+1], q1.val[3], y, 1);
isum[2*iy+0] = vdotq_laneq_s32(isum[2*iy+0], q2.val[0], y, 2);
isum[2*iy+1] = vdotq_laneq_s32(isum[2*iy+1], q2.val[1], y, 2);
isum[2*iy+0] = vdotq_laneq_s32(isum[2*iy+0], q2.val[2], y, 3);
isum[2*iy+1] = vdotq_laneq_s32(isum[2*iy+1], q2.val[3], y, 3);
}
}
for (int iy = 0; iy < nrc_y; ++iy) {
auto d8 = vdupq_n_f32(q8.scale(iy, ibl));
const float * bsum = (const float *)q8.y[iy][ibl].bsums;
auto m8 = vdupq_n_f32(-128.f*bsum[0]);
acc[2*iy+0] = vfmaq_f32(acc[2*iy+0], vmulq_f32(d4l, d8), vcvtq_f32_s32(isum[2*iy+0]));
acc[2*iy+1] = vfmaq_f32(acc[2*iy+1], vmulq_f32(d4h, d8), vcvtq_f32_s32(isum[2*iy+1]));
acc[2*iy+0] = vfmaq_f32(acc[2*iy+0], d4l, m8);
acc[2*iy+1] = vfmaq_f32(acc[2*iy+1], d4l, m8);
}
}
for (int iy = 0; iy < nrc_y; ++iy) {
info.store(ix+0, iy, acc[2*iy+0]);
info.store(ix+4, iy, acc[2*iy+1]);
acc[2*iy+0] = acc[2*iy+1] = vdupq_n_f32(0.f);
}
}
}
void mul_mat_iq4_nl_r4_q8_0_1(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) {
GGML_ASSERT(nrc_x%4 == 0);
Q8<1, block_q8_0_x4> q8(info);
@ -9575,6 +9710,10 @@ bool MulMat::prepare(int typeA, int typeB, int ne00, MulMat& m, int /*Ny*/) {
SET_MUL_MAT_FUNCTIONS(m, mul_mat_q6_k_r4_q8_k);
expected_Btype = GGML_TYPE_Q8_K;
break;
case GGML_TYPE_Q8_K_R8:
SET_MUL_MAT_FUNCTIONS(m, mul_mat_q8_k_r8_q8_k);
expected_Btype = GGML_TYPE_Q8_KR8;
break;
case GGML_TYPE_IQ4_K_R4:
SET_MUL_MAT_FUNCTIONS(m, mul_mat_iq4_k_r4_q8_k);
expected_Btype = GGML_TYPE_Q8_K;

103
ggml/src/iqk/iqk_quantize.cpp
View File

@ -2469,7 +2469,7 @@ size_t quantize_iq6_k(const float * src, void * dst, int64_t nrows, int64_t n_pe
return nrows * nblock * sizeof(block_iq6_k);
}
template <bool is_K32>
template <int q8_type>
void iqk_quantize_row_q8_K_T(const float * x, void * vy, int64_t k) {
assert(k % QK_K == 0);
const int nb = k / QK_K;
@ -2505,7 +2505,7 @@ void iqk_quantize_row_q8_K_T(const float * x, void * vy, int64_t k) {
__m256i i1 = _mm256_cvtps_epi32(v1);
__m256i i2 = _mm256_cvtps_epi32(v2);
__m256i i3 = _mm256_cvtps_epi32(v3);
if constexpr (is_K32) {
if constexpr (q8_type > 0) {
int bsum = hsum_i32_8(_mm256_add_epi32(_mm256_add_epi32(i0, i1), _mm256_add_epi32(i2, i3)));
auto bs = (float *)y[i].bsums;
bs[ib] = d*bsum;
@ -2520,6 +2520,12 @@ void iqk_quantize_row_q8_K_T(const float * x, void * vy, int64_t k) {
_mm256_storeu_si256((__m256i *)q8, i0);
q8 += 32;
}
if constexpr (q8_type == 2) {
auto bs = (float *)y[i].bsums;
float sum = 0;
for (int ib = 0; ib < QK_K/32; ++ib) sum += bs[ib];
bs[0] = sum;
}
}
#else
for (int i = 0; i < nb; i++) {
@ -2545,15 +2551,20 @@ void iqk_quantize_row_q8_K_T(const float * x, void * vy, int64_t k) {
int v = nearest_int(iscale*x[j]);
y[i].qs[j] = MIN(127, v);
}
if constexpr (is_K32) {
if constexpr (q8_type > 0) {
auto bs = (float *)y[i].bsums;
float d = 1/iscale;
float sum = 0;
for (int j = 0; j < QK_K/32; ++j) {
int sum = 0;
for (int ii = 0; ii < 32; ++ii) {
sum += y[i].qs[j*32 + ii];
}
bs[j] = d*sum;
sum += bs[j];
}
if constexpr (q8_type == 2) {
bs[0] = sum;
}
} else {
for (int j = 0; j < QK_K/16; ++j) {
@ -2572,11 +2583,15 @@ void iqk_quantize_row_q8_K_T(const float * x, void * vy, int64_t k) {
}
void iqk_quantize_row_q8_K(const float * x, void * vy, int64_t k) {
iqk_quantize_row_q8_K_T<false>(x, vy, k);
iqk_quantize_row_q8_K_T<0>(x, vy, k);
}
void quantize_row_q8_K32(const float * x, void * vy, int64_t k) {
iqk_quantize_row_q8_K_T<true>(x, vy, k);
iqk_quantize_row_q8_K_T<1>(x, vy, k);
}
void quantize_row_q8_KR8(const float * x, void * vy, int64_t k) {
iqk_quantize_row_q8_K_T<2>(x, vy, k);
}
namespace {
@ -4666,3 +4681,81 @@ void vec_dot_iq4_k_r4_q8_k(int n, float * s, size_t bs, const void * vx, size_t
GGML_UNUSED(by);
}
//
// ========================================= q8_k_r8
//
void quantize_row_q8_k_r8_ref(const float * x, block_q8_k_r8 * y, int64_t k) {
quantize_q8_k_r8(x, (void *)y, 8, k/8, nullptr);
}
void quantize_row_q8_k_r8(const float * x, void * y, int64_t k) {
quantize_q8_k_r8(x, y, 8, k/8, nullptr);
}
static void repack_q8_k(int nrows, int n_per_row, const block_q8_K * x, block_q8_k_r8 * y) {
GGML_ASSERT(nrows%8 == 0);
GGML_ASSERT(n_per_row%QK_K == 0);
int nblock = n_per_row/QK_K;
const block_q8_K * x8[8];
for (int row = 0; row < nrows; row += 8) {
for (int k = 0; k < 8; ++k) x8[k] = x + nblock*k;
for (int ibl = 0; ibl < nblock; ++ibl) {
for (int k = 0; k < 8; ++k) {
y[ibl].d[k] = GGML_FP32_TO_FP16(x8[k][ibl].d);
for (int ib = 0; ib < QK_K/4; ++ib) {
for (int i = 0; i < 4; ++i) y[ibl].qs[32*ib + 4*k + i] = x8[k][ibl].qs[4*ib+i];
}
}
}
x += 4*nblock;
y += nblock;
}
}
size_t quantize_q8_k_r8(const float * src, void * dst, int64_t nrows, int64_t n_per_row, [[maybe_unused]] const float * imatrix) {
GGML_ASSERT(nrows%8 == 0);
GGML_ASSERT(n_per_row%QK_K == 0);
char * qcur = (char *)dst;
auto row_size_0 = ggml_row_size(GGML_TYPE_Q8_K, n_per_row);
auto row_size_1 = ggml_row_size(GGML_TYPE_Q8_K_R8, n_per_row);
std::vector<char> qtmp(8*row_size_0);
for (int row = 0; row < nrows; row += 8) {
quantize_row_q8_K32(src, (void *)qtmp.data(), 8*n_per_row);
repack_q8_k(8, n_per_row, (const block_q8_K *)qtmp.data(), (block_q8_k_r8 *)qcur);
qcur += 8*row_size_1;
src += 8*n_per_row;
}
return nrows*row_size_1;
}
void dequantize_row_q8_k_r8(const block_q8_k_r8 * x, float * y, int64_t k) {
auto n_per_row = k/8;
float * y8[8];
for (int k = 0; k < 8; ++k) y8[k] = y + n_per_row*k;
int nblock = n_per_row/QK_K;
for (int ibl = 0; ibl < nblock; ++ibl) {
for (int k = 0; k < 8; ++k) {
const float d = GGML_FP16_TO_FP32(x[ibl].d[k]);
for (int ib = 0; ib < QK_K/4; ++ib) {
for (int i = 0; i < 4; ++i) {
y8[k][QK_K*ibl+4*ib+i] = d * x[ibl].qs[32*ib+4*k+i];
}
}
}
}
}
void vec_dot_q8_k_r8_q8_k(int n, float * s, size_t bs, const void * vx, size_t bx, const void * vy, size_t by, int nrc) {
#if GGML_USE_IQK_MULMAT
if (iqk_mul_mat(1, 1, n, GGML_TYPE_Q8_K_R8, vx, 0, GGML_TYPE_Q8_K, vy, 0, s, 0, 0, 1)) {
return;
}
#endif
GGML_ASSERT(n%QK4_NL == 0);
GGML_ASSERT(nrc == 1);
GGML_UNUSED(bs);
GGML_UNUSED(bx);
GGML_UNUSED(by);
}

7
ggml/src/iqk/iqk_quantize.h
View File

@ -145,11 +145,18 @@ size_t quantize_iq4_k_r4(const float * GGML_RESTRICT src, void * GGML_RESTRICT d
void dequantize_row_iq4_k_r4(const block_iq4_k_r4 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
void vec_dot_iq4_k_r4_q8_k(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
void quantize_row_q8_k_r8_ref(const float * GGML_RESTRICT x, block_q8_k_r8 * GGML_RESTRICT y, int64_t k);
void quantize_row_q8_k_r8(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
size_t quantize_q8_k_r8(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
void dequantize_row_q8_k_r8(const block_q8_k_r8 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
void vec_dot_q8_k_r8_q8_k(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
void iqk_quantize_row_q8_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
void quantize_row_q8_K64_ref(const float * GGML_RESTRICT x, block_q8_K64 * GGML_RESTRICT y, int64_t k);
void quantize_row_q8_K64(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
void quantize_row_q8_K16(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
void quantize_row_q8_K32(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
void quantize_row_q8_KR8(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
#ifdef __cplusplus
}

1
include/llama.h
View File

@ -193,6 +193,7 @@ extern "C" {
LLAMA_FTYPE_MOSTLY_Q6_0_R4 = 335, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ2_BN_R4 = 337, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ4_K_R4 = 340, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q8_K_R8 = 399, // except 1d tensors
LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
};

16
src/llama.cpp
View File

@ -3843,6 +3843,7 @@ struct llama_model_loader {
case GGML_TYPE_Q5_K_R4: ftype = LLAMA_FTYPE_MOSTLY_Q5_K_R4; break;
case GGML_TYPE_Q6_K: ftype = LLAMA_FTYPE_MOSTLY_Q6_K; break;
case GGML_TYPE_Q6_K_R4: ftype = LLAMA_FTYPE_MOSTLY_Q6_K_R4; break;
case GGML_TYPE_Q8_K_R8: ftype = LLAMA_FTYPE_MOSTLY_Q8_K_R8; break;
case GGML_TYPE_IQ2_XXS: ftype = LLAMA_FTYPE_MOSTLY_IQ2_XXS; break;
case GGML_TYPE_IQ2_XS: ftype = LLAMA_FTYPE_MOSTLY_IQ2_XS; break;
case GGML_TYPE_IQ2_KS: ftype = LLAMA_FTYPE_MOSTLY_IQ2_KS; break;
@ -4560,6 +4561,7 @@ static std::string llama_model_ftype_name(llama_ftype ftype) {
case LLAMA_FTYPE_MOSTLY_Q5_K_M: return "Q5_K - Medium";
case LLAMA_FTYPE_MOSTLY_Q6_K: return "Q6_K";
case LLAMA_FTYPE_MOSTLY_Q6_K_R4: return "Q6_K_R4";
case LLAMA_FTYPE_MOSTLY_Q8_K_R8: return "Q8_K_R8";
case LLAMA_FTYPE_MOSTLY_IQ2_XXS: return "IQ2_XXS - 2.0625 bpw";
case LLAMA_FTYPE_MOSTLY_IQ2_XS: return "IQ2_XS - 2.3125 bpw";
case LLAMA_FTYPE_MOSTLY_IQ2_KS: return "IQ2_KS - 2.1875 bpw";
@ -15766,7 +15768,8 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
ftype == LLAMA_FTYPE_MOSTLY_IQ4_KS || ftype == LLAMA_FTYPE_MOSTLY_IQ4_KSS) && !qs.has_output) {
new_type = GGML_TYPE_IQ5_K;
}
else if (new_type != GGML_TYPE_Q8_0 && new_type != GGML_TYPE_Q8_0_R4 && new_type != GGML_TYPE_IQ6_K && new_type != GGML_TYPE_Q6_K_R4) {
else if (new_type != GGML_TYPE_Q8_0 && new_type != GGML_TYPE_Q8_0_R4 && new_type != GGML_TYPE_IQ6_K && new_type != GGML_TYPE_Q6_K_R4 &&
new_type != GGML_TYPE_Q8_K_R8) {
new_type = GGML_TYPE_Q6_K;
}
}
@ -15812,6 +15815,9 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
else if (new_type == GGML_TYPE_Q6_K_R4) {
new_type = GGML_TYPE_Q6_K;
}
else if (new_type == GGML_TYPE_Q8_K_R8) {
new_type = GGML_TYPE_Q8_0;
}
else if (new_type == GGML_TYPE_IQ4_K_R4) {
new_type = GGML_TYPE_IQ4_K;
}
@ -16099,7 +16105,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
new_type == GGML_TYPE_IQ6_K || new_type == GGML_TYPE_IQ4_KS || new_type == GGML_TYPE_IQ4_XS_R4 ||
new_type == GGML_TYPE_IQ2_KS || new_type == GGML_TYPE_IQ4_KSS || new_type == GGML_TYPE_Q6_K_R4 ||
new_type == GGML_TYPE_Q5_K_R4 || new_type == GGML_TYPE_Q3_K_R4 || new_type == GGML_TYPE_Q2_K_R4 ||
new_type == GGML_TYPE_IQ4_K_R4) {
new_type == GGML_TYPE_IQ4_K_R4|| new_type == GGML_TYPE_Q8_K_R8) {
int nx = tensor->ne[0];
int ny = tensor->ne[1];
if (nx % QK_K != 0) {
@ -16144,6 +16150,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
case GGML_TYPE_Q5_K: new_type = GGML_TYPE_Q6_0; break;
case GGML_TYPE_IQ6_K:
case GGML_TYPE_Q6_K_R4:
case GGML_TYPE_Q8_K_R8:
case GGML_TYPE_Q6_K: new_type = GGML_TYPE_Q8_0; break;
default: throw std::runtime_error("\nUnsupported tensor size encountered\n");
}
@ -16240,6 +16247,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
case LLAMA_FTYPE_MOSTLY_Q5_K_R4: default_type = GGML_TYPE_Q5_K_R4; break;
case LLAMA_FTYPE_MOSTLY_Q6_K: default_type = GGML_TYPE_Q6_K; break;
case LLAMA_FTYPE_MOSTLY_Q6_K_R4: default_type = GGML_TYPE_Q6_K_R4; break;
case LLAMA_FTYPE_MOSTLY_Q8_K_R8: default_type = GGML_TYPE_Q8_K_R8; break;
case LLAMA_FTYPE_MOSTLY_IQ2_XXS: default_type = GGML_TYPE_IQ2_XXS; break;
case LLAMA_FTYPE_MOSTLY_IQ2_XS: default_type = GGML_TYPE_IQ2_XS; break;
case LLAMA_FTYPE_MOSTLY_IQ2_KS: default_type = GGML_TYPE_IQ2_KS; break;
@ -16660,6 +16668,10 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
if (tensor->ne[1] % 4 != 0) new_type = GGML_TYPE_Q6_K;
else chunk_size_multiplier = 4;
}
else if (new_type == GGML_TYPE_Q8_K_R8) {
if (tensor->ne[1] % 8 != 0) new_type = GGML_TYPE_Q8_0;
else chunk_size_multiplier = 8;
}
else if (new_type == GGML_TYPE_IQ2_BN_R4) {
if (tensor->ne[1] % 4 != 0) new_type = GGML_TYPE_IQ2_BN;
else chunk_size_multiplier = 4;