// Auto-generated file. Do not edit! // Template: src/qs8-gemm/rvv.c.in // Generator: tools/xngen // // Copyright 2024 SiFive, Inc. // // 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/gemm.h" #include "xnnpack/math.h" void xnn_qd8_f32_qc8w_gemm_minmax_ukernel_7x4v__rvv( size_t mr, size_t nc, size_t kc, const int8_t* restrict a, size_t a_stride, const void* restrict w, float* restrict c, size_t cm_stride, size_t cn_stride, const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS(1)], const struct xnn_qd8_quantization_params quantization_params[restrict XNN_MIN_ELEMENTS(1)]) { assert(mr != 0); assert(mr <= 7); assert(nc != 0); assert(kc != 0); const int8_t* a0 = a; float* c0 = c; const int8_t* a1 = (const int8_t*) ((uintptr_t) a0 + a_stride); float* c1 = (float*) ((uintptr_t) c0 + cm_stride); if XNN_UNPREDICTABLE(mr < 2) { a1 = a0; c1 = c0; } const int8_t* a2 = (const int8_t*) ((uintptr_t) a1 + a_stride); float* c2 = (float*) ((uintptr_t) c1 + cm_stride); if XNN_UNPREDICTABLE(mr <= 2) { a2 = a1; c2 = c1; } const int8_t* a3 = (const int8_t*) ((uintptr_t) a2 + a_stride); float* c3 = (float*) ((uintptr_t) c2 + cm_stride); if XNN_UNPREDICTABLE(mr < 4) { a3 = a2; c3 = c2; } const int8_t* a4 = (const int8_t*) ((uintptr_t) a3 + a_stride); float* c4 = (float*) ((uintptr_t) c3 + cm_stride); if XNN_UNPREDICTABLE(mr <= 4) { a4 = a3; c4 = c3; } const int8_t* a5 = (const int8_t*) ((uintptr_t) a4 + a_stride); float* c5 = (float*) ((uintptr_t) c4 + cm_stride); if XNN_UNPREDICTABLE(mr < 6) { a5 = a4; c5 = c4; } const int8_t* a6 = (const int8_t*) ((uintptr_t) a5 + a_stride); float* c6 = (float*) ((uintptr_t) c5 + cm_stride); if XNN_UNPREDICTABLE(mr <= 6) { a6 = a5; c6 = c5; } const size_t nr = __riscv_vsetvlmax_e32m4(); size_t vl = nr; do { if XNN_UNLIKELY(nc < nr) { vl = __riscv_vsetvl_e32m4(nc); } nc = nc - vl; vint32m4_t vksum = __riscv_vle32_v_i32m4((const int32_t*)w, vl); w = (const int32_t*) w + nr; const int32_t vinput_zero_point0 = quantization_params[0].zero_point; const int32_t vinput_zero_point1 = quantization_params[1].zero_point; const int32_t vinput_zero_point2 = quantization_params[2].zero_point; const int32_t vinput_zero_point3 = quantization_params[3].zero_point; const int32_t vinput_zero_point4 = quantization_params[4].zero_point; const int32_t vinput_zero_point5 = quantization_params[5].zero_point; const int32_t vinput_zero_point6 = quantization_params[6].zero_point; vint32m4_t vacc0 = __riscv_vmul_vx_i32m4(vksum, vinput_zero_point0, vl); vint32m4_t vacc1 = __riscv_vmul_vx_i32m4(vksum, vinput_zero_point1, vl); vint32m4_t vacc2 = __riscv_vmul_vx_i32m4(vksum, vinput_zero_point2, vl); vint32m4_t vacc3 = __riscv_vmul_vx_i32m4(vksum, vinput_zero_point3, vl); vint32m4_t vacc4 = __riscv_vmul_vx_i32m4(vksum, vinput_zero_point4, vl); vint32m4_t vacc5 = __riscv_vmul_vx_i32m4(vksum, vinput_zero_point5, vl); vint32m4_t vacc6 = __riscv_vmul_vx_i32m4(vksum, vinput_zero_point6, vl); size_t k = kc; do { const int8_t va0 = *a0++; const int8_t va1 = *a1++; const int8_t va2 = *a2++; const int8_t va3 = *a3++; const int8_t va4 = *a4++; const int8_t va5 = *a5++; const int8_t va6 = *a6++; const vint8m1_t vb = __riscv_vle8_v_i8m1((const int8_t*) w, vl); w = (const int8_t*) w + nr; vint16m2_t va0b = __riscv_vwmul_vx_i16m2(vb, va0, vl); vacc0 = __riscv_vwadd_wv_i32m4(vacc0, va0b, vl); vint16m2_t va1b = __riscv_vwmul_vx_i16m2(vb, va1, vl); vacc1 = __riscv_vwadd_wv_i32m4(vacc1, va1b, vl); vint16m2_t va2b = __riscv_vwmul_vx_i16m2(vb, va2, vl); vacc2 = __riscv_vwadd_wv_i32m4(vacc2, va2b, vl); vint16m2_t va3b = __riscv_vwmul_vx_i16m2(vb, va3, vl); vacc3 = __riscv_vwadd_wv_i32m4(vacc3, va3b, vl); vint16m2_t va4b = __riscv_vwmul_vx_i16m2(vb, va4, vl); vacc4 = __riscv_vwadd_wv_i32m4(vacc4, va4b, vl); vint16m2_t va5b = __riscv_vwmul_vx_i16m2(vb, va5, vl); vacc5 = __riscv_vwadd_wv_i32m4(vacc5, va5b, vl); vint16m2_t va6b = __riscv_vwmul_vx_i16m2(vb, va6, vl); vacc6 = __riscv_vwadd_wv_i32m4(vacc6, va6b, vl); k -= sizeof(int8_t); } while (k != 0); // i32 -> f32 vfloat32m4_t vout0 = __riscv_vfcvt_f_x_v_f32m4(vacc0, vl); vfloat32m4_t vout1 = __riscv_vfcvt_f_x_v_f32m4(vacc1, vl); vfloat32m4_t vout2 = __riscv_vfcvt_f_x_v_f32m4(vacc2, vl); vfloat32m4_t vout3 = __riscv_vfcvt_f_x_v_f32m4(vacc3, vl); vfloat32m4_t vout4 = __riscv_vfcvt_f_x_v_f32m4(vacc4, vl); vfloat32m4_t vout5 = __riscv_vfcvt_f_x_v_f32m4(vacc5, vl); vfloat32m4_t vout6 = __riscv_vfcvt_f_x_v_f32m4(vacc6, vl); // vout * input_scale const float vinput_scale0 = quantization_params[0].inv_scale; const float vinput_scale1 = quantization_params[1].inv_scale; const float vinput_scale2 = quantization_params[2].inv_scale; const float vinput_scale3 = quantization_params[3].inv_scale; const float vinput_scale4 = quantization_params[4].inv_scale; const float vinput_scale5 = quantization_params[5].inv_scale; const float vinput_scale6 = quantization_params[6].inv_scale; vout0 = __riscv_vfmul_vf_f32m4(vout0, vinput_scale0, vl); vout1 = __riscv_vfmul_vf_f32m4(vout1, vinput_scale1, vl); vout2 = __riscv_vfmul_vf_f32m4(vout2, vinput_scale2, vl); vout3 = __riscv_vfmul_vf_f32m4(vout3, vinput_scale3, vl); vout4 = __riscv_vfmul_vf_f32m4(vout4, vinput_scale4, vl); vout5 = __riscv_vfmul_vf_f32m4(vout5, vinput_scale5, vl); vout6 = __riscv_vfmul_vf_f32m4(vout6, vinput_scale6, vl); const vfloat32m4_t vfilter_output_scale = __riscv_vle32_v_f32m4((const float*) w, vl); w = (const float*) w + nr; vout0 = __riscv_vfmul_vv_f32m4(vout0, vfilter_output_scale, vl); vout1 = __riscv_vfmul_vv_f32m4(vout1, vfilter_output_scale, vl); vout2 = __riscv_vfmul_vv_f32m4(vout2, vfilter_output_scale, vl); vout3 = __riscv_vfmul_vv_f32m4(vout3, vfilter_output_scale, vl); vout4 = __riscv_vfmul_vv_f32m4(vout4, vfilter_output_scale, vl); vout5 = __riscv_vfmul_vv_f32m4(vout5, vfilter_output_scale, vl); vout6 = __riscv_vfmul_vv_f32m4(vout6, vfilter_output_scale, vl); const vfloat32m4_t vbias = __riscv_vle32_v_f32m4((const float*) w, vl); w = (const float*) w + nr; vout0 = __riscv_vfadd_vv_f32m4(vout0, vbias, vl); vout1 = __riscv_vfadd_vv_f32m4(vout1, vbias, vl); vout2 = __riscv_vfadd_vv_f32m4(vout2, vbias, vl); vout3 = __riscv_vfadd_vv_f32m4(vout3, vbias, vl); vout4 = __riscv_vfadd_vv_f32m4(vout4, vbias, vl); vout5 = __riscv_vfadd_vv_f32m4(vout5, vbias, vl); vout6 = __riscv_vfadd_vv_f32m4(vout6, vbias, vl); const float vmin = params->scalar.min; vout0 = __riscv_vfmax_vf_f32m4(vout0, vmin, vl); vout1 = __riscv_vfmax_vf_f32m4(vout1, vmin, vl); vout2 = __riscv_vfmax_vf_f32m4(vout2, vmin, vl); vout3 = __riscv_vfmax_vf_f32m4(vout3, vmin, vl); vout4 = __riscv_vfmax_vf_f32m4(vout4, vmin, vl); vout5 = __riscv_vfmax_vf_f32m4(vout5, vmin, vl); vout6 = __riscv_vfmax_vf_f32m4(vout6, vmin, vl); const float vmax = params->scalar.max; vout0 = __riscv_vfmin_vf_f32m4(vout0, vmax, vl); vout1 = __riscv_vfmin_vf_f32m4(vout1, vmax, vl); vout2 = __riscv_vfmin_vf_f32m4(vout2, vmax, vl); vout3 = __riscv_vfmin_vf_f32m4(vout3, vmax, vl); vout4 = __riscv_vfmin_vf_f32m4(vout4, vmax, vl); vout5 = __riscv_vfmin_vf_f32m4(vout5, vmax, vl); vout6 = __riscv_vfmin_vf_f32m4(vout6, vmax, vl); // store 7 x vl results to c __riscv_vse32_v_f32m4(c0, vout0, vl); c0 = (float*) ((uintptr_t) c0 + cn_stride); __riscv_vse32_v_f32m4(c1, vout1, vl); c1 = (float*) ((uintptr_t) c1 + cn_stride); __riscv_vse32_v_f32m4(c2, vout2, vl); c2 = (float*) ((uintptr_t) c2 + cn_stride); __riscv_vse32_v_f32m4(c3, vout3, vl); c3 = (float*) ((uintptr_t) c3 + cn_stride); __riscv_vse32_v_f32m4(c4, vout4, vl); c4 = (float*) ((uintptr_t) c4 + cn_stride); __riscv_vse32_v_f32m4(c5, vout5, vl); c5 = (float*) ((uintptr_t) c5 + cn_stride); __riscv_vse32_v_f32m4(c6, vout6, vl); c6 = (float*) ((uintptr_t) c6 + cn_stride); a0 = (const int8_t*) ((uintptr_t) a0 - kc); a1 = (const int8_t*) ((uintptr_t) a1 - kc); a2 = (const int8_t*) ((uintptr_t) a2 - kc); a3 = (const int8_t*) ((uintptr_t) a3 - kc); a4 = (const int8_t*) ((uintptr_t) a4 - kc); a5 = (const int8_t*) ((uintptr_t) a5 - kc); a6 = (const int8_t*) ((uintptr_t) a6 - kc); } while (nc != 0); }