diff options
Diffstat (limited to 'chapters/image.adoc')
| -rw-r--r-- | chapters/image.adoc | 42 |
1 files changed, 24 insertions, 18 deletions
diff --git a/chapters/image.adoc b/chapters/image.adoc index 0b25369..690480c 100644 --- a/chapters/image.adoc +++ b/chapters/image.adoc @@ -81,41 +81,45 @@ ERROR_IF(OH != idiv_check((IH-1)*scale_y_n - offset_y + border_y, scale_y_d) + 1 ERROR_IF(OW != idiv_check((IW-1)*scale_x_n - offset_x + border_x, scale_x_d) + 1); for_each(0 <= n < N, 0 <= oy < OH, 0 <= ox < OW; 0 <= c < C) { out_t acc; - y = oy * scale_y_d + offset_y; - x = ox * scale_x_d + offset_x; - iy = floor(y / scale_y_n); - ix = floor(x / scale_x_n); - if (resize_t == float_t) { - dy = ((float_t)y / (float_t)scale_y_n) - iy; - dx = ((float_t)x / (float_t)scale_x_n) - ix; + resize_t dx, dy; + + int32_t y = oy * scale_y_d + offset_y; + int32_t x = ox * scale_x_d + offset_x; + int16_t iy = floor(y / scale_y_n); + int16_t ix = floor(x / scale_x_n); + + if (is_floating_point(resize_t)) { + dy = ((resize_t)y / (resize_t)scale_y_n) - iy; + dx = ((resize_t)x / (resize_t)scale_x_n) - ix; } else { dy = y - iy * scale_y_n; dx = y - ix * scale_x_n; } // Note that -1 <= iy < IH and -1 <= ix < IW - iy0 = apply_max(iy, 0); - iy1 = apply_min(iy+1, IH-1); - ix0 = apply_max(ix, 0); - ix1 = apply_min(ix+1, IW-1); + int16_t iy0 = apply_max(iy, 0); + int16_t iy1 = apply_min(iy+1, IH-1); + int16_t ix0 = apply_max(ix, 0); + int16_t ix1 = apply_min(ix+1, IW-1); if (mode==BILINEAR) { - v00 = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy0,ix0,c]); - v01 = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy0,ix1,c]); - v10 = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy1,ix0,c]); - v11 = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy1,ix1,c]); + in_t v00 = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy0,ix0,c]); + in_t v01 = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy0,ix1,c]); + in_t v10 = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy1,ix0,c]); + in_t v11 = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy1,ix1,c]); acc = v00 * (scale_y_n - dy) * (scale_x_n - dx); acc += v01 * (scale_y_n - dy) * dx; acc += v10 * dy * (scale_x_n - dx); acc += v11 * dy * dx; tensor_write<out_t>(output, [N,OH,OW,C], [n,oy,ox,c], acc); } else if (mode==NEAREST) { - if (resize_t == float_t) { + int32_t iy, ix; + if (is_floating_point(resize_t)) { iy = (dy >= 0.5) ? iy1 : iy0; ix = (dx >= 0.5) ? ix1 : ix0; } else { iy = (2*dy >= scale_y_n) ? iy1 : iy0; ix = (2*dx >= scale_x_n) ? ix1 : ix0; } - v = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy,ix,c]); + in_t v = tensor_read<in_t>(input, [N,IH,IW,C], [n,iy,ix,c]); tensor_write<out_t>(output, [N,OH,OW,C], [n,oy,ox,c], v); } } @@ -130,7 +134,9 @@ for_each(0 <= n < N, 0 <= oy < OH, 0 <= ox < OW; 0 <= c < C) { |Any|signed 8, nearest |int16_t|int8_t|int8_t |Any|signed 16, bilinear|int16_t|int16_t|int48_t |Any|signed 16, nearest |int16_t|int16_t|int16_t -|MI,MT|floating-point |float_t|float_t|float_t +|MI,MT|fp16|fp32_t|fp16_t|fp16_t +|MI,MT|bf16|fp32_t|bf16_t|bf16_t +|MI,MT|fp32|fp32_t|fp32_t|fp32_t |=== *Resize Modes:* |