ref: f9dc411d89eed99d7def7de1e9dddba782c1212c
parent: e3ce12cfc1c2d2cc245e1a6d49eaf3ff18538547
author: Luc Trudeau <[email protected]>
date: Wed Jun 13 11:24:54 EDT 2018
[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
--- 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, \
@@ -39,12 +31,8 @@
uint32x4_t v_sad = vec_zeros_u32; \
\
do { \- v_a = vec_vsx_ld(0, a); \
- v_b = vec_vsx_ld(0, b); \
+ PROCESS16(0) \
\
- v_abs = vec_sub(vec_max(v_a, v_b), vec_min(v_a, v_b)); \
- v_sad = vec_sum4s(v_abs, v_sad); \
- \
a += a_stride; \
b += b_stride; \
y++; \
@@ -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);