From db3cd886b8303a8a5aff20c76660ed49c2365f94 Mon Sep 17 00:00:00 2001 From: Matthias Vogelgesang Date: Sat, 23 May 2020 20:04:38 +0200 Subject: backproject: unroll for navi10 With that an RX 5600 XT achieves ~22 slices/s according to the benchmark protocol from https://tomopedia.github.io/software/ufo. --- src/kernels/backproject.cl | 3 +++ 1 file changed, 3 insertions(+) diff --git a/src/kernels/backproject.cl b/src/kernels/backproject.cl index 4ee84de..8a81790 100644 --- a/src/kernels/backproject.cl +++ b/src/kernels/backproject.cl @@ -81,6 +81,9 @@ backproject_tex (read_only image2d_t sinogram, #endif #ifdef DEVICE_QUADRO_M6000 #pragma unroll 2 +#endif +#ifdef DEVICE_GFX1010 +#pragma unroll 4 #endif for(int proj = 0; proj < n_projections; proj++) { float h = by * sin_lut[angle_offset + proj] + bx * cos_lut[angle_offset + proj] + axis_pos; -- cgit v1.2.1