diff options
| author | Luc Trudeau <luc@trud.ca> | 2018-06-13 15:24:54 -0400 |
|---|---|---|
| committer | Luc Trudeau <luc@trud.ca> | 2018-06-14 01:57:05 +0000 |
| commit | f9dc411d89eed99d7def7de1e9dddba782c1212c (patch) | |
| tree | 33508fcc22b758f36e855f8c9b7a4f41d57eb7ea /vpx_dsp | |
| parent | e3ce12cfc1c2d2cc245e1a6d49eaf3ff18538547 (diff) | |
| download | libvpx-f9dc411d89eed99d7def7de1e9dddba782c1212c.tar libvpx-f9dc411d89eed99d7def7de1e9dddba782c1212c.tar.gz libvpx-f9dc411d89eed99d7def7de1e9dddba782c1212c.tar.bz2 libvpx-f9dc411d89eed99d7def7de1e9dddba782c1212c.zip | |
[VSX] Optimize PROCESS16 macro
The PROCESS16 macro now uses 8-bit lanes instead of 16-bit lanes.
SADTest Speed Test (POWER8 Model 2.1)
16x8 Old VSX time = 16.7 ms, new VSX time = 9.1 ms [1.8x]
16x16 Old VSX time = 15.7 ms, new VSX time = 7.9 ms [2.0x]
16x32 Old VSX time = 14.4 ms, new VSX time = 7.2 ms [2.0x]
32x16 Old VSX time = 14.0 ms, new VSX time = 7.4 ms [1.9x]
32x32 Old VSX time = 13.4 ms, new VSX time = 6.5 ms [2.0x]
32x64 Old VSX time = 12.7 ms, new VSX time = 6.3 ms [2.0x]
64x32 Old VSX time = 12.6 ms, new VSX time = 6.3 ms [2.0x]
64x64 Old VSX time = 12.7 ms, new VSX time = 6.2 ms [2.0x]
Change-Id: I51776f0e428162e78edde8eac47f30ffd2379873
Diffstat (limited to 'vpx_dsp')
| -rw-r--r-- | vpx_dsp/ppc/sad_vsx.c | 72 |
1 files changed, 27 insertions, 45 deletions
diff --git a/vpx_dsp/ppc/sad_vsx.c b/vpx_dsp/ppc/sad_vsx.c index b41c05b48..18673f682 100644 --- a/vpx_dsp/ppc/sad_vsx.c +++ b/vpx_dsp/ppc/sad_vsx.c @@ -17,19 +17,11 @@ #include "vpx/vpx_integer.h" #include "vpx_ports/mem.h" -#define PROCESS16(offset) \ - v_a = vec_vsx_ld(offset, a); \ - v_b = vec_vsx_ld(offset, b); \ - v_ah = unpack_to_s16_h(v_a); \ - v_al = unpack_to_s16_l(v_a); \ - v_bh = unpack_to_s16_h(v_b); \ - v_bl = unpack_to_s16_l(v_b); \ - v_subh = vec_sub(v_ah, v_bh); \ - v_subl = vec_sub(v_al, v_bl); \ - v_absh = vec_abs(v_subh); \ - v_absl = vec_abs(v_subl); \ - v_sad = vec_sum4s(v_absh, v_sad); \ - v_sad = vec_sum4s(v_absl, v_sad); +#define PROCESS16(offset) \ + v_a = vec_vsx_ld(offset, a); \ + v_b = vec_vsx_ld(offset, b); \ + v_abs = vec_sub(vec_max(v_a, v_b), vec_min(v_a, v_b)); \ + v_sad = vec_sum4s(v_abs, v_sad); #define SAD8(height) \ unsigned int vpx_sad8x##height##_vsx(const uint8_t *a, int a_stride, \ @@ -39,11 +31,7 @@ uint32x4_t v_sad = vec_zeros_u32; \ \ do { \ - v_a = vec_vsx_ld(0, a); \ - v_b = vec_vsx_ld(0, b); \ - \ - v_abs = vec_sub(vec_max(v_a, v_b), vec_min(v_a, v_b)); \ - v_sad = vec_sum4s(v_abs, v_sad); \ + PROCESS16(0) \ \ a += a_stride; \ b += b_stride; \ @@ -56,54 +44,48 @@ #define SAD16(height) \ unsigned int vpx_sad16x##height##_vsx(const uint8_t *a, int a_stride, \ const uint8_t *b, int b_stride) { \ - int y; \ - unsigned int sad[4]; \ - uint8x16_t v_a, v_b; \ - int16x8_t v_ah, v_al, v_bh, v_bl, v_absh, v_absl, v_subh, v_subl; \ - int32x4_t v_sad = vec_splat_s32(0); \ + int y = 0; \ + uint8x16_t v_a, v_b, v_abs; \ + uint32x4_t v_sad = vec_zeros_u32; \ \ - for (y = 0; y < height; y++) { \ + do { \ PROCESS16(0); \ \ a += a_stride; \ b += b_stride; \ - } \ - vec_vsx_st((uint32x4_t)v_sad, 0, sad); \ + y++; \ + } while (y < height); \ \ - return sad[3] + sad[2] + sad[1] + sad[0]; \ + return v_sad[3] + v_sad[2] + v_sad[1] + v_sad[0]; \ } #define SAD32(height) \ unsigned int vpx_sad32x##height##_vsx(const uint8_t *a, int a_stride, \ const uint8_t *b, int b_stride) { \ - int y; \ - unsigned int sad[4]; \ - uint8x16_t v_a, v_b; \ - int16x8_t v_ah, v_al, v_bh, v_bl, v_absh, v_absl, v_subh, v_subl; \ - int32x4_t v_sad = vec_splat_s32(0); \ + int y = 0; \ + uint8x16_t v_a, v_b, v_abs; \ + uint32x4_t v_sad = vec_zeros_u32; \ \ - for (y = 0; y < height; y++) { \ + do { \ PROCESS16(0); \ PROCESS16(16); \ \ a += a_stride; \ b += b_stride; \ - } \ - vec_vsx_st((uint32x4_t)v_sad, 0, sad); \ + y++; \ + } while (y < height); \ \ - return sad[3] + sad[2] + sad[1] + sad[0]; \ + return v_sad[3] + v_sad[2] + v_sad[1] + v_sad[0]; \ } #define SAD64(height) \ unsigned int vpx_sad64x##height##_vsx(const uint8_t *a, int a_stride, \ const uint8_t *b, int b_stride) { \ - int y; \ - unsigned int sad[4]; \ - uint8x16_t v_a, v_b; \ - int16x8_t v_ah, v_al, v_bh, v_bl, v_absh, v_absl, v_subh, v_subl; \ - int32x4_t v_sad = vec_splat_s32(0); \ + int y = 0; \ + uint8x16_t v_a, v_b, v_abs; \ + uint32x4_t v_sad = vec_zeros_u32; \ \ - for (y = 0; y < height; y++) { \ + do { \ PROCESS16(0); \ PROCESS16(16); \ PROCESS16(32); \ @@ -111,10 +93,10 @@ \ a += a_stride; \ b += b_stride; \ - } \ - vec_vsx_st((uint32x4_t)v_sad, 0, sad); \ + y++; \ + } while (y < height); \ \ - return sad[3] + sad[2] + sad[1] + sad[0]; \ + return v_sad[3] + v_sad[2] + v_sad[1] + v_sad[0]; \ } SAD8(4); |