Much faster prompt processing for IQ1_S and IQ1_M on ARM_NEON (#553)

* iq1_s

66.3 t/s -> 168.8 t/s.

* iq1_m

19 t/s -> 163 t/s.

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>

2 files changed, 158 insertions, 2 deletions

diff --git a/ggml/src/iqk/iqk_gemm_1bit.cpp b/ggml/src/iqk/iqk_gemm_1bit.cpp
index 770fbf2c..2c0a1bda 100644
--- a/ggml/src/iqk/iqk_gemm_1bit.cpp
+++ b/ggml/src/iqk/iqk_gemm_1bit.cpp
@@ -2515,6 +2515,154 @@ void mul_mat_iq1_s_q8_K(int n, const void * vx, size_t bx, const DataInfo& info,
         }
     }
 }
+
+inline float convert_to_q8_k_r8(float d0, const int8x16x2_t * qx, const int8_t * scales, uint32_t * block, uint32_t * q8_k) {
+    auto max_i16 = vdupq_n_u16(0);
+    int16x8x4_t q[8];
+    for (int ib32 = 0; ib32 < 8; ++ib32) {
+        auto scale_l = vdup_n_s8(scales[2*ib32+0]);
+        auto scale_h = vdup_n_s8(scales[2*ib32+1]);
+        q[ib32].val[0] = vmull_s8(scale_l, vget_low_s8 (qx[ib32].val[0]));
+        q[ib32].val[1] = vmull_s8(scale_l, vget_high_s8(qx[ib32].val[0]));
+        q[ib32].val[2] = vmull_s8(scale_h, vget_low_s8 (qx[ib32].val[1]));
+        q[ib32].val[3] = vmull_s8(scale_h, vget_high_s8(qx[ib32].val[1]));
+        max_i16 = vmaxq_u16(max_i16, vmaxq_u16(vabsq_s16(q[ib32].val[0]), vabsq_s16(q[ib32].val[1])));
+        max_i16 = vmaxq_u16(max_i16, vmaxq_u16(vabsq_s16(q[ib32].val[2]), vabsq_s16(q[ib32].val[3])));
+    }
+    uint16_t imax = vmaxvq_u16(max_i16);
+    if (!imax) {
+        for (int ib32 = 0; ib32 < 8; ++ib32) for (int l = 0; l < 8; ++l) q8_k[64*ib32 + 8*l] = 0;
+        return 0.f;
+    }
+    float dnew = float(imax) * d0;
+    //auto max_u32 = vmaxq_u32(vmovl_u16(vget_low_u16(max_i16)), vmovl_u16(vget_high_u16(max_i16)));
+    //auto max_f32 = vcvtq_f32_u32(max_u32);
+    //auto dnew = vmaxvq_f32(max_f32) * d0;
+    bool needs_scaling = true;
+    if (dnew <= 1.f) {
+        dnew = 1.f; needs_scaling = false;
+    }
+    auto scale = vdupq_n_f32(1/dnew);
+    for (int ib32 = 0; ib32 < 8; ++ib32) {
+        if (needs_scaling) {
+            for (int l = 0; l < 4; ++l) {
+                auto i1 = vcvtnq_s32_f32(vmulq_f32(scale, vcvtq_f32_s32(vmovl_s16(vget_low_s16 (q[ib32].val[l])))));
+                auto i2 = vcvtnq_s32_f32(vmulq_f32(scale, vcvtq_f32_s32(vmovl_s16(vget_high_s16(q[ib32].val[l])))));
+                q[ib32].val[l] = vcombine_s16(vmovn_s32(i1), vmovn_s32(i2));
+            }
+        }
+        for (int l = 0; l < 2; ++l) {
+            auto s8 = vcombine_s8(vmovn_s16(q[ib32].val[2*l+0]), vmovn_s16(q[ib32].val[2*l+1]));
+            vst1q_s8((int8_t *)block + 16*l, s8);
+        }
+        auto qb = q8_k + 64*ib32;
+        for (int l = 0; l < 8; ++l) {
+            qb[8*l] = block[l];
+        }
+    }
+    return dnew;
+}
+
+void iqk_convert_iq1_s_q8_k_r8(int n, const void * vx, size_t bx, void * vy, int nrc_x) {
+    GGML_ASSERT(n%QK_K == 0);
+    GGML_ASSERT(nrc_x%8 == 0);
+
+    int nb = n/QK_K;
+
+    const block_iq1_s * x8[8];
+
+    block_q8_k_r8 * y = (block_q8_k_r8 *)vy;
+
+    int8_t ls[16];
+
+    uint32_t block[8];
+
+    int8x16x2_t qx[8];
+
+    for (int ix = 0; ix < nrc_x; ix += 8) {
+        for (int k = 0; k < 8; ++k) x8[k] = (const block_iq1_s *)((const char *)vx + (ix + k)*bx);
+        for (int i = 0; i < nb; ++i) {
+            for (int k = 0; k < 8; ++k) {
+                float d = 0.125f * GGML_FP16_TO_FP32(x8[k][i].d);
+                auto qs = x8[k][i].qs;
+                auto qh = x8[k][i].qh;
+                int8x16x2_t value;
+                for (int ib32 = 0; ib32 < 8; ++ib32) {
+                    ls[2*ib32 + 0] = ls[2*ib32 + 1] = (2*((qh[ib32] >> 12) & 7) + 1);
+                    value.val[0] = vreinterpretq_s8_u64(uint64x2_t{iq1s_grid[qs[0] | ((qh[ib32] << 8) & 0x700)], iq1s_grid[qs[1] | ((qh[ib32] << 5) & 0x700)]});
+                    value.val[1] = vreinterpretq_s8_u64(uint64x2_t{iq1s_grid[qs[2] | ((qh[ib32] << 2) & 0x700)], iq1s_grid[qs[3] | ((qh[ib32] >> 1) & 0x700)]});
+                    value.val[0] = vshlq_n_s8(vaddq_s8(value.val[0], vdupq_n_s8(1)), 3);
+                    value.val[1] = vshlq_n_s8(vaddq_s8(value.val[1], vdupq_n_s8(1)), 3);
+                    auto delta = vdupq_n_s8(qh[ib32] & 0x8000 ? -9 : -7);
+                    qx[ib32].val[0] = vaddq_s8(value.val[0], delta);
+                    qx[ib32].val[1] = vaddq_s8(value.val[1], delta);
+                    qs += 4;
+                }
+                float dnew = convert_to_q8_k_r8(1.f/126, qx, ls, block, (uint32_t *)y[i].qs + k);
+                y[i].d[k] = GGML_FP32_TO_FP16(d*dnew);
+            }
+        }
+        y += nb;
+    }
+}
+
+void iqk_convert_iq1_m_q8_k_r8(int n, const void * vx, size_t bx, void * vy, int nrc_x) {
+    GGML_ASSERT(n%QK_K == 0);
+    GGML_ASSERT(nrc_x%8 == 0);
+
+    int nb = n/QK_K;
+
+    const block_iq1_m * x8[8];
+
+    block_q8_k_r8 * y = (block_q8_k_r8 *)vy;
+
+    int8_t ls[16];
+
+    uint32_t block[8];
+
+    int8x16x2_t qx[8];
+
+    uint32x4x2_t mask = {uint32x4_t{0x00000008, 0x00000008, 0x00000080, 0x00000080}, uint32x4_t {0x00080000, 0x00080000, 0x00800000, 0x00800000}};
+
+    for (int ix = 0; ix < nrc_x; ix += 8) {
+        for (int k = 0; k < 8; ++k) x8[k] = (const block_iq1_m *)((const char *)vx + (ix + k)*bx);
+        for (int i = 0; i < nb; ++i) {
+            for (int k = 0; k < 8; ++k) {
+                const uint16_t * sc = (const uint16_t *)x8[k][i].scales;
+                iq1m_scale_t scale;
+                scale.u16 = (sc[0] >> 12) | ((sc[1] >> 8) & 0x00f0) | ((sc[2] >> 4) & 0x0f00) | (sc[3] & 0xf000);
+                float d = 0.125f * GGML_FP16_TO_FP32(scale.f16);
+                auto qs = x8[k][i].qs;
+                auto qh = x8[k][i].qh;
+                int8x16x2_t value;
+                for (int ib32 = 0; ib32 < 8; ++ib32) {
+                    ls[2*ib32 + 0] = (2*((sc[ib32/2] >> (6*(ib32%2)+0)) & 0x7) + 1);
+                    ls[2*ib32 + 1] = (2*((sc[ib32/2] >> (6*(ib32%2)+3)) & 0x7) + 1);
+                    //value = _mm256_set_epi64x(iq1s_grid[qs[3] | ((qh[1] << 4) & 0x700)], iq1s_grid[qs[2] | ((qh[1] << 8) & 0x700)],
+                    //                          iq1s_grid[qs[1] | ((qh[0] << 4) & 0x700)], iq1s_grid[qs[0] | ((qh[0] << 8) & 0x700)]);
+                    value.val[0] = vreinterpretq_s8_u64(uint64x2_t{iq1s_grid[qs[0] | ((qh[0] << 8) & 0x700)], iq1s_grid[qs[1] | ((qh[0] << 4) & 0x700)]});
+                    value.val[1] = vreinterpretq_s8_u64(uint64x2_t{iq1s_grid[qs[2] | ((qh[1] << 8) & 0x700)], iq1s_grid[qs[3] | ((qh[1] << 4) & 0x700)]});
+                    value.val[0] = vshlq_n_s8(vaddq_s8(value.val[0], vdupq_n_s8(1)), 3);
+                    value.val[1] = vshlq_n_s8(vaddq_s8(value.val[1], vdupq_n_s8(1)), 3);
+
+                    auto aux = vdupq_n_u32(qh[0] | qh[1] << 16);
+                    uint32x4x2_t delta_mask{ vceqq_u32(vandq_u32(aux, mask.val[0]), mask.val[0]), vceqq_u32(vandq_u32(aux, mask.val[1]), mask.val[1]) };
+                    uint8x16x2_t delta{ vaddq_s8(vdupq_n_s8(7), vandq_s8(vdupq_n_s8(2), vreinterpretq_s8_u32(delta_mask.val[0]))),
+                                        vaddq_s8(vdupq_n_s8(7), vandq_s8(vdupq_n_s8(2), vreinterpretq_s8_u32(delta_mask.val[1]))) };
+                    qx[ib32].val[0] = vsubq_s8(value.val[0], delta.val[0]);
+                    qx[ib32].val[1] = vsubq_s8(value.val[1], delta.val[1]);
+
+                    qs += 4;
+                    qh += 2;
+                }
+                float dnew = convert_to_q8_k_r8(1.f/126, qx, ls, block, (uint32_t *)y[i].qs + k);
+                y[i].d[k] = GGML_FP32_TO_FP16(d*dnew);
+            }
+        }
+        y += nb;
+    }
+}
+
 }
 
 bool iqk_set_kernels_1bit(int ne00, int typeA, int typeB, std::array<mul_mat_t, IQK_MAX_NY>& funcs, mul_mat_t& func16) {
@@ -2573,8 +2721,14 @@ bool iqk_set_kernels_1bit(int ne00, int typeA, int typeB, std::array<mul_mat_t,
 
 }
 
-bool iqk_convert_1bit_q80_r8([[maybe_unused]] int type, [[maybe_unused]] int n, [[maybe_unused]] const void * vx, [[maybe_unused]] size_t bx, [[maybe_unused]] void * vy, [[maybe_unused]] int nrc_x) {
-    return false;
+bool iqk_convert_1bit_q80_r8(int type, int n, const void * vx, size_t bx, void * vy, int nrc_x) {
+    if (n%QK_K != 0 || nrc_x%8 != 0) return false;
+    switch (ggml_type(type)) {
+        case GGML_TYPE_IQ1_S: iqk_convert_iq1_s_q8_k_r8(n, vx, bx, vy, nrc_x); break;
+        case GGML_TYPE_IQ1_M: iqk_convert_iq1_m_q8_k_r8(n, vx, bx, vy, nrc_x); break;
+        default: return false;
+    }
+    return true;
 }
 
 #endif
diff --git a/ggml/src/iqk/iqk_mul_mat.cpp b/ggml/src/iqk/iqk_mul_mat.cpp
index 8138fa7e..1e015fb5 100644
--- a/ggml/src/iqk/iqk_mul_mat.cpp
+++ b/ggml/src/iqk/iqk_mul_mat.cpp
@@ -276,6 +276,8 @@ struct MulMat {
             case GGML_TYPE_Q4_K   : return nrc_y >= 32 ? GGML_TYPE_Q8_1    : type;
             case GGML_TYPE_Q5_K   : return nrc_y >= 32 ? GGML_TYPE_Q8_1    : type;
             case GGML_TYPE_Q6_K   : return nrc_y >= 64 ? GGML_TYPE_Q8_0_R8 : type;
+            case GGML_TYPE_IQ1_S  : return nrc_y >= 32 ? GGML_TYPE_Q8_K_R8 : type;
+            case GGML_TYPE_IQ1_M  : return nrc_y >=  8 ? GGML_TYPE_Q8_K_R8 : type;
             case GGML_TYPE_IQ2_XXS: return nrc_y >= 32 ? GGML_TYPE_Q8_K_R8 : type;
             case GGML_TYPE_IQ2_XS : return nrc_y >= 32 ? GGML_TYPE_Q8_K_R8 : type;
             case GGML_TYPE_IQ2_S  : return nrc_y >= 32 ? GGML_TYPE_Q8_K_R8 : type;