diff options
author | Ronald S. Bultje <rsbultje@gmail.com> | 2011-02-21 17:07:13 +0300 |
---|---|---|
committer | Ronald S. Bultje <rsbultje@gmail.com> | 2011-02-21 18:23:44 +0300 |
commit | f8bed30d8b176fa030f6737765338bb4a2bcabc9 (patch) | |
tree | f55e179732f7db45c5fcfecd80757676bdbf7fcc /libavcodec/ppc | |
parent | 8d9ac969cb4ac3e3e18f6425703af4d7aec6c513 (diff) |
VC1: merge idct8x8, coeff adjustments and put_pixels.
Merging these functions allows merging some loops, which makes the
results (particularly after SIMD optimizations) much faster.
Diffstat (limited to 'libavcodec/ppc')
-rw-r--r-- | libavcodec/ppc/vc1dsp_altivec.c | 64 |
1 files changed, 61 insertions, 3 deletions
diff --git a/libavcodec/ppc/vc1dsp_altivec.c b/libavcodec/ppc/vc1dsp_altivec.c index 5a0dddbe1d..b5cc71c3cf 100644 --- a/libavcodec/ppc/vc1dsp_altivec.c +++ b/libavcodec/ppc/vc1dsp_altivec.c @@ -130,7 +130,8 @@ do { \ /** Do inverse transform on 8x8 block */ -static void vc1_inv_trans_8x8_altivec(DCTELEM block[64]) +static void vc1_inv_trans_8x8_altivec(DCTELEM block[64], + int sign, int rangered) { vector signed short src0, src1, src2, src3, src4, src5, src6, src7; vector signed int s0, s1, s2, s3, s4, s5, s6, s7; @@ -144,7 +145,9 @@ static void vc1_inv_trans_8x8_altivec(DCTELEM block[64]) const vector unsigned int vec_2 = vec_splat_u32(2); const vector signed int vec_1s = vec_splat_s32(1); const vector unsigned int vec_1 = vec_splat_u32(1); - + const vector unsigned short rangered_shift = vec_splat_u16(1); + const vector signed short signed_bias = vec_sl(vec_splat_u16(4), + vec_splat_u16(4)); src0 = vec_ld( 0, block); src1 = vec_ld( 16, block); @@ -214,6 +217,27 @@ static void vc1_inv_trans_8x8_altivec(DCTELEM block[64]) src6 = vec_pack(sE, s6); src7 = vec_pack(sF, s7); + if (rangered) { + if (!sign) { + vec_sub(src0, signed_bias); + vec_sub(src1, signed_bias); + vec_sub(src2, signed_bias); + vec_sub(src3, signed_bias); + vec_sub(src4, signed_bias); + vec_sub(src5, signed_bias); + vec_sub(src6, signed_bias); + vec_sub(src7, signed_bias); + } + vec_sl(src0, rangered_shift); + vec_sl(src1, rangered_shift); + vec_sl(src2, rangered_shift); + vec_sl(src3, rangered_shift); + vec_sl(src4, rangered_shift); + vec_sl(src5, rangered_shift); + vec_sl(src6, rangered_shift); + vec_sl(src7, rangered_shift); + } + vec_st(src0, 0, block); vec_st(src1, 16, block); vec_st(src2, 32, block); @@ -224,6 +248,36 @@ static void vc1_inv_trans_8x8_altivec(DCTELEM block[64]) vec_st(src7,112, block); } +static void vc1_inv_trans_8x8_add_altivec(uint8_t *dest, int stride, DCTELEM *b) +{ + vc1_inv_trans_8x8_altivec(b, 0, 0); + ff_add_pixels_clamped_c(b, dest, stride); +} + +static void vc1_inv_trans_8x8_put_signed_altivec(uint8_t *dest, int stride, DCTELEM *b) +{ + vc1_inv_trans_8x8_altivec(b, 1, 0); + ff_put_signed_pixels_clamped_c(b, dest, stride); +} + +static void vc1_inv_trans_8x8_put_signed_rangered_altivec(uint8_t *dest, int stride, DCTELEM *b) +{ + vc1_inv_trans_8x8_altivec(b, 1, 1); + ff_put_signed_pixels_clamped_c(b, dest, stride); +} + +static void vc1_inv_trans_8x8_put_altivec(uint8_t *dest, int stride, DCTELEM *b) +{ + vc1_inv_trans_8x8_altivec(b, 0, 0); + ff_put_pixels_clamped_c(b, dest, stride); +} + +static void vc1_inv_trans_8x8_put_rangered_altivec(uint8_t *dest, int stride, DCTELEM *b) +{ + vc1_inv_trans_8x8_altivec(b, 0, 1); + ff_put_pixels_clamped_c(b, dest, stride); +} + /** Do inverse transform on 8x4 part of block */ static void vc1_inv_trans_8x4_altivec(uint8_t *dest, int stride, DCTELEM *block) @@ -342,7 +396,11 @@ void ff_vc1dsp_init_altivec(VC1DSPContext* dsp) if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) return; - dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_altivec; + dsp->vc1_inv_trans_8x8_add = vc1_inv_trans_8x8_add_altivec; + dsp->vc1_inv_trans_8x8_put_signed[0] = vc1_inv_trans_8x8_put_signed_altivec; + dsp->vc1_inv_trans_8x8_put_signed[1] = vc1_inv_trans_8x8_put_signed_rangered_altivec; + dsp->vc1_inv_trans_8x8_put[0] = vc1_inv_trans_8x8_put_altivec; + dsp->vc1_inv_trans_8x8_put[1] = vc1_inv_trans_8x8_put_rangered_altivec; dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_altivec; dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_altivec; dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_altivec; |