Mercurial > libavcodec.hg
diff ppc/h264_template_altivec.c @ 7333:a8a79f5385f6 libavcodec
cosmetics: Reformat PPC code in libavcodec according to style guidelines.
This includes indentation changes, comment reformatting, consistent brace
placement and some prettyprinting.
| author | diego |
|---|---|
| date | Sun, 20 Jul 2008 18:58:30 +0000 |
| parents | 292269939c50 |
| children | 266d4949aa15 |
line wrap: on
line diff
--- a/ppc/h264_template_altivec.c Sun Jul 20 18:06:41 2008 +0000 +++ b/ppc/h264_template_altivec.c Sun Jul 20 18:58:30 2008 +0000 @@ -206,489 +206,489 @@ /* this code assume stride % 16 == 0 */ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) { - POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_h_lowpass_num, 1); - register int i; + POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_h_lowpass_num, 1); + register int i; - LOAD_ZERO; - const vec_u8_t permM2 = vec_lvsl(-2, src); - const vec_u8_t permM1 = vec_lvsl(-1, src); - const vec_u8_t permP0 = vec_lvsl(+0, src); - const vec_u8_t permP1 = vec_lvsl(+1, src); - const vec_u8_t permP2 = vec_lvsl(+2, src); - const vec_u8_t permP3 = vec_lvsl(+3, src); - const vec_s16_t v5ss = vec_splat_s16(5); - const vec_u16_t v5us = vec_splat_u16(5); - const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); - const vec_s16_t v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4)); + LOAD_ZERO; + const vec_u8_t permM2 = vec_lvsl(-2, src); + const vec_u8_t permM1 = vec_lvsl(-1, src); + const vec_u8_t permP0 = vec_lvsl(+0, src); + const vec_u8_t permP1 = vec_lvsl(+1, src); + const vec_u8_t permP2 = vec_lvsl(+2, src); + const vec_u8_t permP3 = vec_lvsl(+3, src); + const vec_s16_t v5ss = vec_splat_s16(5); + const vec_u16_t v5us = vec_splat_u16(5); + const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); + const vec_s16_t v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4)); - vec_u8_t srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; + vec_u8_t srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; - register int align = ((((unsigned long)src) - 2) % 16); + register int align = ((((unsigned long)src) - 2) % 16); - vec_s16_t srcP0A, srcP0B, srcP1A, srcP1B, - srcP2A, srcP2B, srcP3A, srcP3B, - srcM1A, srcM1B, srcM2A, srcM2B, - sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, - pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, - psumA, psumB, sumA, sumB; + vec_s16_t srcP0A, srcP0B, srcP1A, srcP1B, + srcP2A, srcP2B, srcP3A, srcP3B, + srcM1A, srcM1B, srcM2A, srcM2B, + sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, + pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, + psumA, psumB, sumA, sumB; - vec_u8_t sum, vdst, fsum; + vec_u8_t sum, vdst, fsum; - POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1); + POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1); - for (i = 0 ; i < 16 ; i ++) { - vec_u8_t srcR1 = vec_ld(-2, src); - vec_u8_t srcR2 = vec_ld(14, src); + for (i = 0 ; i < 16 ; i ++) { + vec_u8_t srcR1 = vec_ld(-2, src); + vec_u8_t srcR2 = vec_ld(14, src); - switch (align) { - default: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = vec_perm(srcR1, srcR2, permP3); - } break; - case 11: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = srcR2; - } break; - case 12: { - vec_u8_t srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = srcR2; - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 13: { - vec_u8_t srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = srcR2; - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 14: { - vec_u8_t srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = srcR2; - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 15: { - vec_u8_t srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = srcR2; - srcP0 = vec_perm(srcR2, srcR3, permP0); - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; + switch (align) { + default: { + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = vec_perm(srcR1, srcR2, permP0); + srcP1 = vec_perm(srcR1, srcR2, permP1); + srcP2 = vec_perm(srcR1, srcR2, permP2); + srcP3 = vec_perm(srcR1, srcR2, permP3); + } break; + case 11: { + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = vec_perm(srcR1, srcR2, permP0); + srcP1 = vec_perm(srcR1, srcR2, permP1); + srcP2 = vec_perm(srcR1, srcR2, permP2); + srcP3 = srcR2; + } break; + case 12: { + vec_u8_t srcR3 = vec_ld(30, src); + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = vec_perm(srcR1, srcR2, permP0); + srcP1 = vec_perm(srcR1, srcR2, permP1); + srcP2 = srcR2; + srcP3 = vec_perm(srcR2, srcR3, permP3); + } break; + case 13: { + vec_u8_t srcR3 = vec_ld(30, src); + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = vec_perm(srcR1, srcR2, permP0); + srcP1 = srcR2; + srcP2 = vec_perm(srcR2, srcR3, permP2); + srcP3 = vec_perm(srcR2, srcR3, permP3); + } break; + case 14: { + vec_u8_t srcR3 = vec_ld(30, src); + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = srcR2; + srcP1 = vec_perm(srcR2, srcR3, permP1); + srcP2 = vec_perm(srcR2, srcR3, permP2); + srcP3 = vec_perm(srcR2, srcR3, permP3); + } break; + case 15: { + vec_u8_t srcR3 = vec_ld(30, src); + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = srcR2; + srcP0 = vec_perm(srcR2, srcR3, permP0); + srcP1 = vec_perm(srcR2, srcR3, permP1); + srcP2 = vec_perm(srcR2, srcR3, permP2); + srcP3 = vec_perm(srcR2, srcR3, permP3); + } break; + } + + srcP0A = (vec_s16_t) vec_mergeh(zero_u8v, srcP0); + srcP0B = (vec_s16_t) vec_mergel(zero_u8v, srcP0); + srcP1A = (vec_s16_t) vec_mergeh(zero_u8v, srcP1); + srcP1B = (vec_s16_t) vec_mergel(zero_u8v, srcP1); + + srcP2A = (vec_s16_t) vec_mergeh(zero_u8v, srcP2); + srcP2B = (vec_s16_t) vec_mergel(zero_u8v, srcP2); + srcP3A = (vec_s16_t) vec_mergeh(zero_u8v, srcP3); + srcP3B = (vec_s16_t) vec_mergel(zero_u8v, srcP3); + + srcM1A = (vec_s16_t) vec_mergeh(zero_u8v, srcM1); + srcM1B = (vec_s16_t) vec_mergel(zero_u8v, srcM1); + srcM2A = (vec_s16_t) vec_mergeh(zero_u8v, srcM2); + srcM2B = (vec_s16_t) vec_mergel(zero_u8v, srcM2); + + sum1A = vec_adds(srcP0A, srcP1A); + sum1B = vec_adds(srcP0B, srcP1B); + sum2A = vec_adds(srcM1A, srcP2A); + sum2B = vec_adds(srcM1B, srcP2B); + sum3A = vec_adds(srcM2A, srcP3A); + sum3B = vec_adds(srcM2B, srcP3B); + + pp1A = vec_mladd(sum1A, v20ss, v16ss); + pp1B = vec_mladd(sum1B, v20ss, v16ss); + + pp2A = vec_mladd(sum2A, v5ss, zero_s16v); + pp2B = vec_mladd(sum2B, v5ss, zero_s16v); + + pp3A = vec_add(sum3A, pp1A); + pp3B = vec_add(sum3B, pp1B); + + psumA = vec_sub(pp3A, pp2A); + psumB = vec_sub(pp3B, pp2B); + + sumA = vec_sra(psumA, v5us); + sumB = vec_sra(psumB, v5us); + + sum = vec_packsu(sumA, sumB); + + ASSERT_ALIGNED(dst); + vdst = vec_ld(0, dst); + + OP_U8_ALTIVEC(fsum, sum, vdst); + + vec_st(fsum, 0, dst); + + src += srcStride; + dst += dstStride; } - - srcP0A = (vec_s16_t) vec_mergeh(zero_u8v, srcP0); - srcP0B = (vec_s16_t) vec_mergel(zero_u8v, srcP0); - srcP1A = (vec_s16_t) vec_mergeh(zero_u8v, srcP1); - srcP1B = (vec_s16_t) vec_mergel(zero_u8v, srcP1); - - srcP2A = (vec_s16_t) vec_mergeh(zero_u8v, srcP2); - srcP2B = (vec_s16_t) vec_mergel(zero_u8v, srcP2); - srcP3A = (vec_s16_t) vec_mergeh(zero_u8v, srcP3); - srcP3B = (vec_s16_t) vec_mergel(zero_u8v, srcP3); - - srcM1A = (vec_s16_t) vec_mergeh(zero_u8v, srcM1); - srcM1B = (vec_s16_t) vec_mergel(zero_u8v, srcM1); - srcM2A = (vec_s16_t) vec_mergeh(zero_u8v, srcM2); - srcM2B = (vec_s16_t) vec_mergel(zero_u8v, srcM2); - - sum1A = vec_adds(srcP0A, srcP1A); - sum1B = vec_adds(srcP0B, srcP1B); - sum2A = vec_adds(srcM1A, srcP2A); - sum2B = vec_adds(srcM1B, srcP2B); - sum3A = vec_adds(srcM2A, srcP3A); - sum3B = vec_adds(srcM2B, srcP3B); - - pp1A = vec_mladd(sum1A, v20ss, v16ss); - pp1B = vec_mladd(sum1B, v20ss, v16ss); - - pp2A = vec_mladd(sum2A, v5ss, zero_s16v); - pp2B = vec_mladd(sum2B, v5ss, zero_s16v); - - pp3A = vec_add(sum3A, pp1A); - pp3B = vec_add(sum3B, pp1B); - - psumA = vec_sub(pp3A, pp2A); - psumB = vec_sub(pp3B, pp2B); - - sumA = vec_sra(psumA, v5us); - sumB = vec_sra(psumB, v5us); - - sum = vec_packsu(sumA, sumB); - - ASSERT_ALIGNED(dst); - vdst = vec_ld(0, dst); - - OP_U8_ALTIVEC(fsum, sum, vdst); - - vec_st(fsum, 0, dst); - - src += srcStride; - dst += dstStride; - } -POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1); + POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1); } /* this code assume stride % 16 == 0 */ static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) { - POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_v_lowpass_num, 1); + POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_v_lowpass_num, 1); - register int i; + register int i; - LOAD_ZERO; - const vec_u8_t perm = vec_lvsl(0, src); - const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); - const vec_u16_t v5us = vec_splat_u16(5); - const vec_s16_t v5ss = vec_splat_s16(5); - const vec_s16_t v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4)); + LOAD_ZERO; + const vec_u8_t perm = vec_lvsl(0, src); + const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); + const vec_u16_t v5us = vec_splat_u16(5); + const vec_s16_t v5ss = vec_splat_s16(5); + const vec_s16_t v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4)); - uint8_t *srcbis = src - (srcStride * 2); + uint8_t *srcbis = src - (srcStride * 2); - const vec_u8_t srcM2a = vec_ld(0, srcbis); - const vec_u8_t srcM2b = vec_ld(16, srcbis); - const vec_u8_t srcM2 = vec_perm(srcM2a, srcM2b, perm); -// srcbis += srcStride; - const vec_u8_t srcM1a = vec_ld(0, srcbis += srcStride); - const vec_u8_t srcM1b = vec_ld(16, srcbis); - const vec_u8_t srcM1 = vec_perm(srcM1a, srcM1b, perm); -// srcbis += srcStride; - const vec_u8_t srcP0a = vec_ld(0, srcbis += srcStride); - const vec_u8_t srcP0b = vec_ld(16, srcbis); - const vec_u8_t srcP0 = vec_perm(srcP0a, srcP0b, perm); -// srcbis += srcStride; - const vec_u8_t srcP1a = vec_ld(0, srcbis += srcStride); - const vec_u8_t srcP1b = vec_ld(16, srcbis); - const vec_u8_t srcP1 = vec_perm(srcP1a, srcP1b, perm); -// srcbis += srcStride; - const vec_u8_t srcP2a = vec_ld(0, srcbis += srcStride); - const vec_u8_t srcP2b = vec_ld(16, srcbis); - const vec_u8_t srcP2 = vec_perm(srcP2a, srcP2b, perm); -// srcbis += srcStride; + const vec_u8_t srcM2a = vec_ld(0, srcbis); + const vec_u8_t srcM2b = vec_ld(16, srcbis); + const vec_u8_t srcM2 = vec_perm(srcM2a, srcM2b, perm); + //srcbis += srcStride; + const vec_u8_t srcM1a = vec_ld(0, srcbis += srcStride); + const vec_u8_t srcM1b = vec_ld(16, srcbis); + const vec_u8_t srcM1 = vec_perm(srcM1a, srcM1b, perm); + //srcbis += srcStride; + const vec_u8_t srcP0a = vec_ld(0, srcbis += srcStride); + const vec_u8_t srcP0b = vec_ld(16, srcbis); + const vec_u8_t srcP0 = vec_perm(srcP0a, srcP0b, perm); + //srcbis += srcStride; + const vec_u8_t srcP1a = vec_ld(0, srcbis += srcStride); + const vec_u8_t srcP1b = vec_ld(16, srcbis); + const vec_u8_t srcP1 = vec_perm(srcP1a, srcP1b, perm); + //srcbis += srcStride; + const vec_u8_t srcP2a = vec_ld(0, srcbis += srcStride); + const vec_u8_t srcP2b = vec_ld(16, srcbis); + const vec_u8_t srcP2 = vec_perm(srcP2a, srcP2b, perm); + //srcbis += srcStride; - vec_s16_t srcM2ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcM2); - vec_s16_t srcM2ssB = (vec_s16_t) vec_mergel(zero_u8v, srcM2); - vec_s16_t srcM1ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcM1); - vec_s16_t srcM1ssB = (vec_s16_t) vec_mergel(zero_u8v, srcM1); - vec_s16_t srcP0ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP0); - vec_s16_t srcP0ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP0); - vec_s16_t srcP1ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP1); - vec_s16_t srcP1ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP1); - vec_s16_t srcP2ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP2); - vec_s16_t srcP2ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP2); + vec_s16_t srcM2ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcM2); + vec_s16_t srcM2ssB = (vec_s16_t) vec_mergel(zero_u8v, srcM2); + vec_s16_t srcM1ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcM1); + vec_s16_t srcM1ssB = (vec_s16_t) vec_mergel(zero_u8v, srcM1); + vec_s16_t srcP0ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP0); + vec_s16_t srcP0ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP0); + vec_s16_t srcP1ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP1); + vec_s16_t srcP1ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP1); + vec_s16_t srcP2ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP2); + vec_s16_t srcP2ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP2); - vec_s16_t pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, - psumA, psumB, sumA, sumB, - srcP3ssA, srcP3ssB, - sum1A, sum1B, sum2A, sum2B, sum3A, sum3B; + vec_s16_t pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, + psumA, psumB, sumA, sumB, + srcP3ssA, srcP3ssB, + sum1A, sum1B, sum2A, sum2B, sum3A, sum3B; - vec_u8_t sum, vdst, fsum, srcP3a, srcP3b, srcP3; + vec_u8_t sum, vdst, fsum, srcP3a, srcP3b, srcP3; - POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1); + POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1); - for (i = 0 ; i < 16 ; i++) { - srcP3a = vec_ld(0, srcbis += srcStride); - srcP3b = vec_ld(16, srcbis); - srcP3 = vec_perm(srcP3a, srcP3b, perm); - srcP3ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP3); - srcP3ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP3); -// srcbis += srcStride; + for (i = 0 ; i < 16 ; i++) { + srcP3a = vec_ld(0, srcbis += srcStride); + srcP3b = vec_ld(16, srcbis); + srcP3 = vec_perm(srcP3a, srcP3b, perm); + srcP3ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP3); + srcP3ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP3); + //srcbis += srcStride; - sum1A = vec_adds(srcP0ssA, srcP1ssA); - sum1B = vec_adds(srcP0ssB, srcP1ssB); - sum2A = vec_adds(srcM1ssA, srcP2ssA); - sum2B = vec_adds(srcM1ssB, srcP2ssB); - sum3A = vec_adds(srcM2ssA, srcP3ssA); - sum3B = vec_adds(srcM2ssB, srcP3ssB); + sum1A = vec_adds(srcP0ssA, srcP1ssA); + sum1B = vec_adds(srcP0ssB, srcP1ssB); + sum2A = vec_adds(srcM1ssA, srcP2ssA); + sum2B = vec_adds(srcM1ssB, srcP2ssB); + sum3A = vec_adds(srcM2ssA, srcP3ssA); + sum3B = vec_adds(srcM2ssB, srcP3ssB); - srcM2ssA = srcM1ssA; - srcM2ssB = srcM1ssB; - srcM1ssA = srcP0ssA; - srcM1ssB = srcP0ssB; - srcP0ssA = srcP1ssA; - srcP0ssB = srcP1ssB; - srcP1ssA = srcP2ssA; - srcP1ssB = srcP2ssB; - srcP2ssA = srcP3ssA; - srcP2ssB = srcP3ssB; + srcM2ssA = srcM1ssA; + srcM2ssB = srcM1ssB; + srcM1ssA = srcP0ssA; + srcM1ssB = srcP0ssB; + srcP0ssA = srcP1ssA; + srcP0ssB = srcP1ssB; + srcP1ssA = srcP2ssA; + srcP1ssB = srcP2ssB; + srcP2ssA = srcP3ssA; + srcP2ssB = srcP3ssB; - pp1A = vec_mladd(sum1A, v20ss, v16ss); - pp1B = vec_mladd(sum1B, v20ss, v16ss); + pp1A = vec_mladd(sum1A, v20ss, v16ss); + pp1B = vec_mladd(sum1B, v20ss, v16ss); - pp2A = vec_mladd(sum2A, v5ss, zero_s16v); - pp2B = vec_mladd(sum2B, v5ss, zero_s16v); + pp2A = vec_mladd(sum2A, v5ss, zero_s16v); + pp2B = vec_mladd(sum2B, v5ss, zero_s16v); - pp3A = vec_add(sum3A, pp1A); - pp3B = vec_add(sum3B, pp1B); + pp3A = vec_add(sum3A, pp1A); + pp3B = vec_add(sum3B, pp1B); - psumA = vec_sub(pp3A, pp2A); - psumB = vec_sub(pp3B, pp2B); + psumA = vec_sub(pp3A, pp2A); + psumB = vec_sub(pp3B, pp2B); - sumA = vec_sra(psumA, v5us); - sumB = vec_sra(psumB, v5us); + sumA = vec_sra(psumA, v5us); + sumB = vec_sra(psumB, v5us); - sum = vec_packsu(sumA, sumB); + sum = vec_packsu(sumA, sumB); - ASSERT_ALIGNED(dst); - vdst = vec_ld(0, dst); + ASSERT_ALIGNED(dst); + vdst = vec_ld(0, dst); - OP_U8_ALTIVEC(fsum, sum, vdst); + OP_U8_ALTIVEC(fsum, sum, vdst); - vec_st(fsum, 0, dst); + vec_st(fsum, 0, dst); - dst += dstStride; - } - POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1); + dst += dstStride; + } + POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1); } /* this code assume stride % 16 == 0 *and* tmp is properly aligned */ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) { - POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_hv_lowpass_num, 1); - register int i; - LOAD_ZERO; - const vec_u8_t permM2 = vec_lvsl(-2, src); - const vec_u8_t permM1 = vec_lvsl(-1, src); - const vec_u8_t permP0 = vec_lvsl(+0, src); - const vec_u8_t permP1 = vec_lvsl(+1, src); - const vec_u8_t permP2 = vec_lvsl(+2, src); - const vec_u8_t permP3 = vec_lvsl(+3, src); - const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); - const vec_u32_t v10ui = vec_splat_u32(10); - const vec_s16_t v5ss = vec_splat_s16(5); - const vec_s16_t v1ss = vec_splat_s16(1); - const vec_s32_t v512si = vec_sl(vec_splat_s32(1),vec_splat_u32(9)); - const vec_u32_t v16ui = vec_sl(vec_splat_u32(1),vec_splat_u32(4)); + POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_hv_lowpass_num, 1); + register int i; + LOAD_ZERO; + const vec_u8_t permM2 = vec_lvsl(-2, src); + const vec_u8_t permM1 = vec_lvsl(-1, src); + const vec_u8_t permP0 = vec_lvsl(+0, src); + const vec_u8_t permP1 = vec_lvsl(+1, src); + const vec_u8_t permP2 = vec_lvsl(+2, src); + const vec_u8_t permP3 = vec_lvsl(+3, src); + const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); + const vec_u32_t v10ui = vec_splat_u32(10); + const vec_s16_t v5ss = vec_splat_s16(5); + const vec_s16_t v1ss = vec_splat_s16(1); + const vec_s32_t v512si = vec_sl(vec_splat_s32(1),vec_splat_u32(9)); + const vec_u32_t v16ui = vec_sl(vec_splat_u32(1),vec_splat_u32(4)); - register int align = ((((unsigned long)src) - 2) % 16); + register int align = ((((unsigned long)src) - 2) % 16); - vec_s16_t srcP0A, srcP0B, srcP1A, srcP1B, - srcP2A, srcP2B, srcP3A, srcP3B, - srcM1A, srcM1B, srcM2A, srcM2B, - sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, - pp1A, pp1B, pp2A, pp2B, psumA, psumB; + vec_s16_t srcP0A, srcP0B, srcP1A, srcP1B, + srcP2A, srcP2B, srcP3A, srcP3B, + srcM1A, srcM1B, srcM2A, srcM2B, + sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, + pp1A, pp1B, pp2A, pp2B, psumA, psumB; - const vec_u8_t mperm = (const vec_u8_t) - AVV(0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B, - 0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F); - int16_t *tmpbis = tmp; + const vec_u8_t mperm = (const vec_u8_t) + AVV(0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B, + 0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F); + int16_t *tmpbis = tmp; - vec_s16_t tmpM1ssA, tmpM1ssB, tmpM2ssA, tmpM2ssB, - tmpP0ssA, tmpP0ssB, tmpP1ssA, tmpP1ssB, - tmpP2ssA, tmpP2ssB; + vec_s16_t tmpM1ssA, tmpM1ssB, tmpM2ssA, tmpM2ssB, + tmpP0ssA, tmpP0ssB, tmpP1ssA, tmpP1ssB, + tmpP2ssA, tmpP2ssB; - vec_s32_t pp1Ae, pp1Ao, pp1Be, pp1Bo, pp2Ae, pp2Ao, pp2Be, pp2Bo, - pp3Ae, pp3Ao, pp3Be, pp3Bo, pp1cAe, pp1cAo, pp1cBe, pp1cBo, - pp32Ae, pp32Ao, pp32Be, pp32Bo, sumAe, sumAo, sumBe, sumBo, - ssumAe, ssumAo, ssumBe, ssumBo; - vec_u8_t fsum, sumv, sum, vdst; - vec_s16_t ssume, ssumo; + vec_s32_t pp1Ae, pp1Ao, pp1Be, pp1Bo, pp2Ae, pp2Ao, pp2Be, pp2Bo, + pp3Ae, pp3Ao, pp3Be, pp3Bo, pp1cAe, pp1cAo, pp1cBe, pp1cBo, + pp32Ae, pp32Ao, pp32Be, pp32Bo, sumAe, sumAo, sumBe, sumBo, + ssumAe, ssumAo, ssumBe, ssumBo; + vec_u8_t fsum, sumv, sum, vdst; + vec_s16_t ssume, ssumo; - POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1); - src -= (2 * srcStride); - for (i = 0 ; i < 21 ; i ++) { - vec_u8_t srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; - vec_u8_t srcR1 = vec_ld(-2, src); - vec_u8_t srcR2 = vec_ld(14, src); + POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1); + src -= (2 * srcStride); + for (i = 0 ; i < 21 ; i ++) { + vec_u8_t srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; + vec_u8_t srcR1 = vec_ld(-2, src); + vec_u8_t srcR2 = vec_ld(14, src); - switch (align) { - default: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = vec_perm(srcR1, srcR2, permP3); - } break; - case 11: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = srcR2; - } break; - case 12: { - vec_u8_t srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = srcR2; - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 13: { - vec_u8_t srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = srcR2; - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 14: { - vec_u8_t srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = srcR2; - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 15: { - vec_u8_t srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = srcR2; - srcP0 = vec_perm(srcR2, srcR3, permP0); - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; + switch (align) { + default: { + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = vec_perm(srcR1, srcR2, permP0); + srcP1 = vec_perm(srcR1, srcR2, permP1); + srcP2 = vec_perm(srcR1, srcR2, permP2); + srcP3 = vec_perm(srcR1, srcR2, permP3); + } break; + case 11: { + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = vec_perm(srcR1, srcR2, permP0); + srcP1 = vec_perm(srcR1, srcR2, permP1); + srcP2 = vec_perm(srcR1, srcR2, permP2); + srcP3 = srcR2; + } break; + case 12: { + vec_u8_t srcR3 = vec_ld(30, src); + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = vec_perm(srcR1, srcR2, permP0); + srcP1 = vec_perm(srcR1, srcR2, permP1); + srcP2 = srcR2; + srcP3 = vec_perm(srcR2, srcR3, permP3); + } break; + case 13: { + vec_u8_t srcR3 = vec_ld(30, src); + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = vec_perm(srcR1, srcR2, permP0); + srcP1 = srcR2; + srcP2 = vec_perm(srcR2, srcR3, permP2); + srcP3 = vec_perm(srcR2, srcR3, permP3); + } break; + case 14: { + vec_u8_t srcR3 = vec_ld(30, src); + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = vec_perm(srcR1, srcR2, permM1); + srcP0 = srcR2; + srcP1 = vec_perm(srcR2, srcR3, permP1); + srcP2 = vec_perm(srcR2, srcR3, permP2); + srcP3 = vec_perm(srcR2, srcR3, permP3); + } break; + case 15: { + vec_u8_t srcR3 = vec_ld(30, src); + srcM2 = vec_perm(srcR1, srcR2, permM2); + srcM1 = srcR2; + srcP0 = vec_perm(srcR2, srcR3, permP0); + srcP1 = vec_perm(srcR2, srcR3, permP1); + srcP2 = vec_perm(srcR2, srcR3, permP2); + srcP3 = vec_perm(srcR2, srcR3, permP3); + } break; + } + + srcP0A = (vec_s16_t) vec_mergeh(zero_u8v, srcP0); + srcP0B = (vec_s16_t) vec_mergel(zero_u8v, srcP0); + srcP1A = (vec_s16_t) vec_mergeh(zero_u8v, srcP1); + srcP1B = (vec_s16_t) vec_mergel(zero_u8v, srcP1); + + srcP2A = (vec_s16_t) vec_mergeh(zero_u8v, srcP2); + srcP2B = (vec_s16_t) vec_mergel(zero_u8v, srcP2); + srcP3A = (vec_s16_t) vec_mergeh(zero_u8v, srcP3); + srcP3B = (vec_s16_t) vec_mergel(zero_u8v, srcP3); + + srcM1A = (vec_s16_t) vec_mergeh(zero_u8v, srcM1); + srcM1B = (vec_s16_t) vec_mergel(zero_u8v, srcM1); + srcM2A = (vec_s16_t) vec_mergeh(zero_u8v, srcM2); + srcM2B = (vec_s16_t) vec_mergel(zero_u8v, srcM2); + + sum1A = vec_adds(srcP0A, srcP1A); + sum1B = vec_adds(srcP0B, srcP1B); + sum2A = vec_adds(srcM1A, srcP2A); + sum2B = vec_adds(srcM1B, srcP2B); + sum3A = vec_adds(srcM2A, srcP3A); + sum3B = vec_adds(srcM2B, srcP3B); + + pp1A = vec_mladd(sum1A, v20ss, sum3A); + pp1B = vec_mladd(sum1B, v20ss, sum3B); + + pp2A = vec_mladd(sum2A, v5ss, zero_s16v); + pp2B = vec_mladd(sum2B, v5ss, zero_s16v); + + psumA = vec_sub(pp1A, pp2A); + psumB = vec_sub(pp1B, pp2B); + + vec_st(psumA, 0, tmp); + vec_st(psumB, 16, tmp); + + src += srcStride; + tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */ } - srcP0A = (vec_s16_t) vec_mergeh(zero_u8v, srcP0); - srcP0B = (vec_s16_t) vec_mergel(zero_u8v, srcP0); - srcP1A = (vec_s16_t) vec_mergeh(zero_u8v, srcP1); - srcP1B = (vec_s16_t) vec_mergel(zero_u8v, srcP1); - - srcP2A = (vec_s16_t) vec_mergeh(zero_u8v, srcP2); - srcP2B = (vec_s16_t) vec_mergel(zero_u8v, srcP2); - srcP3A = (vec_s16_t) vec_mergeh(zero_u8v, srcP3); - srcP3B = (vec_s16_t) vec_mergel(zero_u8v, srcP3); - - srcM1A = (vec_s16_t) vec_mergeh(zero_u8v, srcM1); - srcM1B = (vec_s16_t) vec_mergel(zero_u8v, srcM1); - srcM2A = (vec_s16_t) vec_mergeh(zero_u8v, srcM2); - srcM2B = (vec_s16_t) vec_mergel(zero_u8v, srcM2); - - sum1A = vec_adds(srcP0A, srcP1A); - sum1B = vec_adds(srcP0B, srcP1B); - sum2A = vec_adds(srcM1A, srcP2A); - sum2B = vec_adds(srcM1B, srcP2B); - sum3A = vec_adds(srcM2A, srcP3A); - sum3B = vec_adds(srcM2B, srcP3B); - - pp1A = vec_mladd(sum1A, v20ss, sum3A); - pp1B = vec_mladd(sum1B, v20ss, sum3B); - - pp2A = vec_mladd(sum2A, v5ss, zero_s16v); - pp2B = vec_mladd(sum2B, v5ss, zero_s16v); - - psumA = vec_sub(pp1A, pp2A); - psumB = vec_sub(pp1B, pp2B); - - vec_st(psumA, 0, tmp); - vec_st(psumB, 16, tmp); - - src += srcStride; - tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */ - } - - tmpM2ssA = vec_ld(0, tmpbis); - tmpM2ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - tmpM1ssA = vec_ld(0, tmpbis); - tmpM1ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - tmpP0ssA = vec_ld(0, tmpbis); - tmpP0ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - tmpP1ssA = vec_ld(0, tmpbis); - tmpP1ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - tmpP2ssA = vec_ld(0, tmpbis); - tmpP2ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - - for (i = 0 ; i < 16 ; i++) { - const vec_s16_t tmpP3ssA = vec_ld(0, tmpbis); - const vec_s16_t tmpP3ssB = vec_ld(16, tmpbis); - - const vec_s16_t sum1A = vec_adds(tmpP0ssA, tmpP1ssA); - const vec_s16_t sum1B = vec_adds(tmpP0ssB, tmpP1ssB); - const vec_s16_t sum2A = vec_adds(tmpM1ssA, tmpP2ssA); - const vec_s16_t sum2B = vec_adds(tmpM1ssB, tmpP2ssB); - const vec_s16_t sum3A = vec_adds(tmpM2ssA, tmpP3ssA); - const vec_s16_t sum3B = vec_adds(tmpM2ssB, tmpP3ssB); - + tmpM2ssA = vec_ld(0, tmpbis); + tmpM2ssB = vec_ld(16, tmpbis); + tmpbis += tmpStride; + tmpM1ssA = vec_ld(0, tmpbis); + tmpM1ssB = vec_ld(16, tmpbis); + tmpbis += tmpStride; + tmpP0ssA = vec_ld(0, tmpbis); + tmpP0ssB = vec_ld(16, tmpbis); + tmpbis += tmpStride; + tmpP1ssA = vec_ld(0, tmpbis); + tmpP1ssB = vec_ld(16, tmpbis); + tmpbis += tmpStride; + tmpP2ssA = vec_ld(0, tmpbis); + tmpP2ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; - tmpM2ssA = tmpM1ssA; - tmpM2ssB = tmpM1ssB; - tmpM1ssA = tmpP0ssA; - tmpM1ssB = tmpP0ssB; - tmpP0ssA = tmpP1ssA; - tmpP0ssB = tmpP1ssB; - tmpP1ssA = tmpP2ssA; - tmpP1ssB = tmpP2ssB; - tmpP2ssA = tmpP3ssA; - tmpP2ssB = tmpP3ssB; + for (i = 0 ; i < 16 ; i++) { + const vec_s16_t tmpP3ssA = vec_ld(0, tmpbis); + const vec_s16_t tmpP3ssB = vec_ld(16, tmpbis); + + const vec_s16_t sum1A = vec_adds(tmpP0ssA, tmpP1ssA); + const vec_s16_t sum1B = vec_adds(tmpP0ssB, tmpP1ssB); + const vec_s16_t sum2A = vec_adds(tmpM1ssA, tmpP2ssA); + const vec_s16_t sum2B = vec_adds(tmpM1ssB, tmpP2ssB); + const vec_s16_t sum3A = vec_adds(tmpM2ssA, tmpP3ssA); + const vec_s16_t sum3B = vec_adds(tmpM2ssB, tmpP3ssB); + + tmpbis += tmpStride; - pp1Ae = vec_mule(sum1A, v20ss); - pp1Ao = vec_mulo(sum1A, v20ss); - pp1Be = vec_mule(sum1B, v20ss); - pp1Bo = vec_mulo(sum1B, v20ss); - - pp2Ae = vec_mule(sum2A, v5ss); - pp2Ao = vec_mulo(sum2A, v5ss); - pp2Be = vec_mule(sum2B, v5ss); - pp2Bo = vec_mulo(sum2B, v5ss); + tmpM2ssA = tmpM1ssA; + tmpM2ssB = tmpM1ssB; + tmpM1ssA = tmpP0ssA; + tmpM1ssB = tmpP0ssB; + tmpP0ssA = tmpP1ssA; + tmpP0ssB = tmpP1ssB; + tmpP1ssA = tmpP2ssA; + tmpP1ssB = tmpP2ssB; + tmpP2ssA = tmpP3ssA; + tmpP2ssB = tmpP3ssB; - pp3Ae = vec_sra((vec_s32_t)sum3A, v16ui); - pp3Ao = vec_mulo(sum3A, v1ss); - pp3Be = vec_sra((vec_s32_t)sum3B, v16ui); - pp3Bo = vec_mulo(sum3B, v1ss); + pp1Ae = vec_mule(sum1A, v20ss); + pp1Ao = vec_mulo(sum1A, v20ss); + pp1Be = vec_mule(sum1B, v20ss); + pp1Bo = vec_mulo(sum1B, v20ss); - pp1cAe = vec_add(pp1Ae, v512si); - pp1cAo = vec_add(pp1Ao, v512si); - pp1cBe = vec_add(pp1Be, v512si); - pp1cBo = vec_add(pp1Bo, v512si); + pp2Ae = vec_mule(sum2A, v5ss); + pp2Ao = vec_mulo(sum2A, v5ss); + pp2Be = vec_mule(sum2B, v5ss); + pp2Bo = vec_mulo(sum2B, v5ss); - pp32Ae = vec_sub(pp3Ae, pp2Ae); - pp32Ao = vec_sub(pp3Ao, pp2Ao); - pp32Be = vec_sub(pp3Be, pp2Be); - pp32Bo = vec_sub(pp3Bo, pp2Bo); + pp3Ae = vec_sra((vec_s32_t)sum3A, v16ui); + pp3Ao = vec_mulo(sum3A, v1ss); + pp3Be = vec_sra((vec_s32_t)sum3B, v16ui); + pp3Bo = vec_mulo(sum3B, v1ss); + + pp1cAe = vec_add(pp1Ae, v512si); + pp1cAo = vec_add(pp1Ao, v512si); + pp1cBe = vec_add(pp1Be, v512si); + pp1cBo = vec_add(pp1Bo, v512si); - sumAe = vec_add(pp1cAe, pp32Ae); - sumAo = vec_add(pp1cAo, pp32Ao); - sumBe = vec_add(pp1cBe, pp32Be); - sumBo = vec_add(pp1cBo, pp32Bo); + pp32Ae = vec_sub(pp3Ae, pp2Ae); + pp32Ao = vec_sub(pp3Ao, pp2Ao); + pp32Be = vec_sub(pp3Be, pp2Be); + pp32Bo = vec_sub(pp3Bo, pp2Bo); - ssumAe = vec_sra(sumAe, v10ui); - ssumAo = vec_sra(sumAo, v10ui); - ssumBe = vec_sra(sumBe, v10ui); - ssumBo = vec_sra(sumBo, v10ui); + sumAe = vec_add(pp1cAe, pp32Ae); + sumAo = vec_add(pp1cAo, pp32Ao); + sumBe = vec_add(pp1cBe, pp32Be); + sumBo = vec_add(pp1cBo, pp32Bo); - ssume = vec_packs(ssumAe, ssumBe); - ssumo = vec_packs(ssumAo, ssumBo); + ssumAe = vec_sra(sumAe, v10ui); + ssumAo = vec_sra(sumAo, v10ui); + ssumBe = vec_sra(sumBe, v10ui); + ssumBo = vec_sra(sumBo, v10ui); - sumv = vec_packsu(ssume, ssumo); - sum = vec_perm(sumv, sumv, mperm); + ssume = vec_packs(ssumAe, ssumBe); + ssumo = vec_packs(ssumAo, ssumBo); - ASSERT_ALIGNED(dst); - vdst = vec_ld(0, dst); + sumv = vec_packsu(ssume, ssumo); + sum = vec_perm(sumv, sumv, mperm); - OP_U8_ALTIVEC(fsum, sum, vdst); + ASSERT_ALIGNED(dst); + vdst = vec_ld(0, dst); - vec_st(fsum, 0, dst); + OP_U8_ALTIVEC(fsum, sum, vdst); - dst += dstStride; - } - POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1); + vec_st(fsum, 0, dst); + + dst += dstStride; + } + POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1); }