// Auto-generated file. Do not edit! // Template: src/f32-igemm/neon-ld64.c.in // Generator: tools/xngen // // Copyright 2019 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include #include #include "xnnpack/igemm.h" void xnn_f32_igemm_minmax_ukernel_4x8__aarch64_neonfma_lane_ld64( size_t mr, size_t nc, size_t kc, size_t ks, const float** restrict a, const float* restrict w, float* restrict c, size_t cm_stride, size_t cn_stride, size_t a_offset, const float* zero, const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) { assert(mr != 0); assert(mr <= 4); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(float) == 0); assert(ks != 0); assert(ks % (4 * sizeof(void*)) == 0); assert(a_offset % sizeof(float) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); float* c0 = c; float* c1 = (float*) ((uintptr_t) c0 + cm_stride); if XNN_UNPREDICTABLE(mr < 2) { c1 = c0; } float* c2 = (float*) ((uintptr_t) c1 + cm_stride); if XNN_UNPREDICTABLE(mr <= 2) { c2 = c1; } float* c3 = (float*) ((uintptr_t) c2 + cm_stride); if XNN_UNPREDICTABLE(mr != 4) { c3 = c2; } do { float32x4_t vacc0x0 = vld1q_f32(w); w += 4; float32x4_t vacc0x1 = vld1q_f32(w); w += 4; float32x4_t vacc1x0 = vacc0x0; float32x4_t vacc1x1 = vacc0x1; float32x4_t vacc2x0 = vacc0x0; float32x4_t vacc2x1 = vacc0x1; float32x4_t vacc3x0 = vacc0x0; float32x4_t vacc3x1 = vacc0x1; size_t p = ks; do { const float* restrict a0 = a[0]; assert(a0 != NULL); if XNN_UNPREDICTABLE(a0 != zero) { a0 = (const float*) ((uintptr_t) a0 + a_offset); } const float* restrict a1 = a[1]; assert(a1 != NULL); if XNN_UNPREDICTABLE(a1 != zero) { a1 = (const float*) ((uintptr_t) a1 + a_offset); } const float* restrict a2 = a[2]; assert(a2 != NULL); if XNN_UNPREDICTABLE(a2 != zero) { a2 = (const float*) ((uintptr_t) a2 + a_offset); } const float* restrict a3 = a[3]; assert(a3 != NULL); if XNN_UNPREDICTABLE(a3 != zero) { a3 = (const float*) ((uintptr_t) a3 + a_offset); } a += 4; size_t k = kc; for (; k >= 2 * sizeof(float); k -= 2 * sizeof(float)) { const float32x2_t va0 = vld1_f32(a0); a0 += 2; const float32x2_t va1 = vld1_f32(a1); a1 += 2; const float32x2_t va2 = vld1_f32(a2); a2 += 2; const float32x2_t va3 = vld1_f32(a3); a3 += 2; const float32x4_t vb0123c0 = vld1q_f32(w); w += 4; const float32x4_t vb4567c0 = vld1q_f32(w); w += 4; vacc0x0 = vfmaq_lane_f32(vacc0x0, vb0123c0, va0, 0); vacc1x0 = vfmaq_lane_f32(vacc1x0, vb0123c0, va1, 0); vacc2x0 = vfmaq_lane_f32(vacc2x0, vb0123c0, va2, 0); vacc3x0 = vfmaq_lane_f32(vacc3x0, vb0123c0, va3, 0); vacc0x1 = vfmaq_lane_f32(vacc0x1, vb4567c0, va0, 0); vacc1x1 = vfmaq_lane_f32(vacc1x1, vb4567c0, va1, 0); vacc2x1 = vfmaq_lane_f32(vacc2x1, vb4567c0, va2, 0); vacc3x1 = vfmaq_lane_f32(vacc3x1, vb4567c0, va3, 0); const float32x4_t vb0123c1 = vld1q_f32(w); w += 4; const float32x4_t vb4567c1 = vld1q_f32(w); w += 4; vacc0x0 = vfmaq_lane_f32(vacc0x0, vb0123c1, va0, 1); vacc1x0 = vfmaq_lane_f32(vacc1x0, vb0123c1, va1, 1); vacc2x0 = vfmaq_lane_f32(vacc2x0, vb0123c1, va2, 1); vacc3x0 = vfmaq_lane_f32(vacc3x0, vb0123c1, va3, 1); vacc0x1 = vfmaq_lane_f32(vacc0x1, vb4567c1, va0, 1); vacc1x1 = vfmaq_lane_f32(vacc1x1, vb4567c1, va1, 1); vacc2x1 = vfmaq_lane_f32(vacc2x1, vb4567c1, va2, 1); vacc3x1 = vfmaq_lane_f32(vacc3x1, vb4567c1, va3, 1); } if XNN_UNLIKELY(k != 0) { const float32x4_t va0 = vld1q_dup_f32(a0); const float32x4_t va1 = vld1q_dup_f32(a1); const float32x4_t va2 = vld1q_dup_f32(a2); const float32x4_t va3 = vld1q_dup_f32(a3); const float32x4_t vb0123 = vld1q_f32(w); w += 4; const float32x4_t vb4567 = vld1q_f32(w); w += 4; vacc0x0 = vfmaq_f32(vacc0x0, va0, vb0123); vacc1x0 = vfmaq_f32(vacc1x0, va1, vb0123); vacc2x0 = vfmaq_f32(vacc2x0, va2, vb0123); vacc3x0 = vfmaq_f32(vacc3x0, va3, vb0123); vacc0x1 = vfmaq_f32(vacc0x1, va0, vb4567); vacc1x1 = vfmaq_f32(vacc1x1, va1, vb4567); vacc2x1 = vfmaq_f32(vacc2x1, va2, vb4567); vacc3x1 = vfmaq_f32(vacc3x1, va3, vb4567); } p -= 4 * sizeof(void*); } while (p != 0); const float32x4_t vmax = vld1q_dup_f32(¶ms->scalar.max); vacc0x0 = vminq_f32(vacc0x0, vmax); vacc1x0 = vminq_f32(vacc1x0, vmax); vacc2x0 = vminq_f32(vacc2x0, vmax); vacc3x0 = vminq_f32(vacc3x0, vmax); vacc0x1 = vminq_f32(vacc0x1, vmax); vacc1x1 = vminq_f32(vacc1x1, vmax); vacc2x1 = vminq_f32(vacc2x1, vmax); vacc3x1 = vminq_f32(vacc3x1, vmax); const float32x4_t vmin = vld1q_dup_f32(¶ms->scalar.min); vacc0x0 = vmaxq_f32(vacc0x0, vmin); vacc1x0 = vmaxq_f32(vacc1x0, vmin); vacc2x0 = vmaxq_f32(vacc2x0, vmin); vacc3x0 = vmaxq_f32(vacc3x0, vmin); vacc0x1 = vmaxq_f32(vacc0x1, vmin); vacc1x1 = vmaxq_f32(vacc1x1, vmin); vacc2x1 = vmaxq_f32(vacc2x1, vmin); vacc3x1 = vmaxq_f32(vacc3x1, vmin); if XNN_LIKELY(nc >= 8) { vst1q_f32(c3, vacc3x0); vst1q_f32(c3 + 4, vacc3x1); c3 = (float*) ((uintptr_t) c3 + cn_stride); vst1q_f32(c2, vacc2x0); vst1q_f32(c2 + 4, vacc2x1); c2 = (float*) ((uintptr_t) c2 + cn_stride); vst1q_f32(c1, vacc1x0); vst1q_f32(c1 + 4, vacc1x1); c1 = (float*) ((uintptr_t) c1 + cn_stride); vst1q_f32(c0, vacc0x0); vst1q_f32(c0 + 4, vacc0x1); c0 = (float*) ((uintptr_t) c0 + cn_stride); a = (const float**restrict) ((uintptr_t) a - ks); nc -= 8; } else { if (nc & 4) { vst1q_f32(c3, vacc3x0); c3 += 4; vst1q_f32(c2, vacc2x0); c2 += 4; vst1q_f32(c1, vacc1x0); c1 += 4; vst1q_f32(c0, vacc0x0); c0 += 4; vacc3x0 = vacc3x1; vacc2x0 = vacc2x1; vacc1x0 = vacc1x1; vacc0x0 = vacc0x1; } float32x2_t vacc3 = vget_low_f32(vacc3x0); float32x2_t vacc2 = vget_low_f32(vacc2x0); float32x2_t vacc1 = vget_low_f32(vacc1x0); float32x2_t vacc0 = vget_low_f32(vacc0x0); if (nc & 2) { vst1_f32(c3, vacc3); c3 += 2; vst1_f32(c2, vacc2); c2 += 2; vst1_f32(c1, vacc1); c1 += 2; vst1_f32(c0, vacc0); c0 += 2; vacc3 = vget_high_f32(vacc3x0); vacc2 = vget_high_f32(vacc2x0); vacc1 = vget_high_f32(vacc1x0); vacc0 = vget_high_f32(vacc0x0); } if (nc & 1) { vst1_lane_f32(c3, vacc3, 0); vst1_lane_f32(c2, vacc2, 0); vst1_lane_f32(c1, vacc1, 0); vst1_lane_f32(c0, vacc0, 0); } nc = 0; } } while (nc != 0); }