eden-emu/opus
20
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Neon: Make gcc actually generate VMLA instructions for sparse mul

Browse source 
Otherwise it was splitting the mla into a mul and an add
This commit is contained in:
Jean-Marc Valin 2019-03-20 12:57:59 -04:00
parent 2bc20e65c7
commit a09815925a
1 changed files with 5 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

10
dnn/vec_neon.h
View file

@ -187,13 +187,13 @@ static void sparse_sgemv_accum16(float *out, const float *w, int rows, const int
for (j=0;j<cols;j++)
{
float xj= x[*idx++];
float32x4_t xj= vld1q_dup_f32(&x[*idx++]);
float32x4_t wvec;
wvec = vld1q_f32(&w[0]); y0_3 = vmlaq_n_f32(y0_3, wvec, xj);
wvec = vld1q_f32(&w[4]); y4_7 = vmlaq_n_f32(y4_7, wvec, xj);
wvec = vld1q_f32(&w[8]); y8_11 = vmlaq_n_f32(y8_11, wvec, xj);
wvec = vld1q_f32(&w[12]); y12_15 = vmlaq_n_f32(y12_15, wvec, xj);
wvec = vld1q_f32(&w[0]); y0_3 = vmlaq_f32(y0_3, wvec, xj);
wvec = vld1q_f32(&w[4]); y4_7 = vmlaq_f32(y4_7, wvec, xj);
wvec = vld1q_f32(&w[8]); y8_11 = vmlaq_f32(y8_11, wvec, xj);
wvec = vld1q_f32(&w[12]); y12_15 = vmlaq_f32(y12_15, wvec, xj);
w += 16;
}