[FFmpeg-devel] [PATCH] avcodec/mips: Improve hevc uni 4 tap hv mc msa functions
Manojkumar Bhosale
Manojkumar.Bhosale at imgtec.com
Tue Nov 7 07:30:02 EET 2017
LGTM
-----Original Message-----
From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On Behalf Of kaustubh.raste at imgtec.com
Sent: Monday, November 6, 2017 4:21 PM
To: ffmpeg-devel at ffmpeg.org
Cc: Kaustubh Raste
Subject: [FFmpeg-devel] [PATCH] avcodec/mips: Improve hevc uni 4 tap hv mc msa functions
From: Kaustubh Raste <kaustubh.raste at imgtec.com>
Use global mask buffer for appropriate mask load.
Remove unused macro and table.
Signed-off-by: Kaustubh Raste <kaustubh.raste at imgtec.com>
---
libavcodec/mips/hevc_mc_uni_msa.c | 1125 +++++++++++++++++++------------------
1 file changed, 566 insertions(+), 559 deletions(-)
diff --git a/libavcodec/mips/hevc_mc_uni_msa.c b/libavcodec/mips/hevc_mc_uni_msa.c
index 993dad0..740c970 100644
--- a/libavcodec/mips/hevc_mc_uni_msa.c
+++ b/libavcodec/mips/hevc_mc_uni_msa.c
@@ -283,25 +283,6 @@ static void copy_width64_msa(uint8_t *src, int32_t src_stride,
}
}
-static const uint8_t mc_filt_mask_arr[16 * 3] = {
- /* 8 width cases */
- 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
- /* 4 width cases */
- 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20,
- /* 4 width cases */
- 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28
-};
-
-#define FILT_4TAP_DPADD_S_H(vec0, vec1, filt0, filt1) \
-( { \
- v8i16 tmp0; \
- \
- tmp0 = __msa_dotp_s_h((v16i8) vec0, (v16i8) filt0); \
- tmp0 = __msa_dpadd_s_h(tmp0, (v16i8) vec1, (v16i8) filt1); \
- \
- tmp0; \
-} )
-
static void common_hz_8t_4x4_msa(uint8_t *src, int32_t src_stride,
uint8_t *dst, int32_t dst_stride,
const int8_t *filter)
@@ -3109,19 +3090,18 @@ static void hevc_hv_uni_4t_4x2_msa(uint8_t *src,
uint8_t *dst,
int32_t dst_stride,
const int8_t *filter_x,
- const int8_t *filter_y,
- int32_t height)
+ const int8_t *filter_y)
{
+ v16u8 out;
v16i8 src0, src1, src2, src3, src4;
v8i16 filt0, filt1;
- v4i32 filt_h0, filt_h1;
- v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+ v8i16 filt_h0, filt_h1;
+ v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
v16i8 mask1;
- v8i16 filter_vec, const_vec;
+ v8i16 filter_vec, tmp;
v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
- v8i16 dst0, dst1, dst2, dst3, dst4;
- v4i32 dst0_r, dst1_r;
- v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+ v8i16 dst20, dst31, dst42, dst10, dst32, dst21, dst43;
+ v4i32 dst0, dst1;
src -= (src_stride + 1);
@@ -3129,60 +3109,35 @@ static void hevc_hv_uni_4t_4x2_msa(uint8_t *src,
SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
filter_vec = LD_SH(filter_y);
- vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
- filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
- SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
mask1 = mask0 + 2;
- const_vec = __msa_ldi_h(128);
- const_vec <<= 6;
-
- LD_SB3(src, src_stride, src0, src1, src2);
- src += (3 * src_stride);
-
- XORI_B3_128_SB(src0, src1, src2);
-
- VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
- VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
- VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
-
- dst0 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
- dst1 = const_vec;
- DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
-
- ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r);
- LD_SB2(src, src_stride, src3, src4);
- XORI_B2_128_SB(src3, src4);
-
- /* row 3 */
- VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
- dst3 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
-
- dst32_r = __msa_ilvr_h(dst3, dst2);
- dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
- dst0_r >>= 6;
-
- /* row 4 */
- VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
- dst4 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
-
- dst43_r = __msa_ilvr_h(dst4, dst3);
- dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
- dst1_r >>= 6;
-
- dst0_r = (v4i32) __msa_pckev_h((v8i16) dst1_r, (v8i16) dst0_r);
- dst0_r = (v4i32) __msa_srari_h((v8i16) dst0_r, 6);
- dst0_r = (v4i32) CLIP_SH_0_255(dst0_r);
- dst0_r = (v4i32) __msa_pckev_b((v16i8) dst0_r, (v16i8) dst0_r);
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+ XORI_B5_128_SB(src0, src1, src2, src3, src4);
- ST4x2_UB(dst0_r, dst, dst_stride);
+ VSHF_B2_SB(src0, src2, src0, src2, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src1, src3, src1, src3, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src2, src4, src2, src4, mask0, mask1, vec4, vec5);
+
+ dst20 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst31 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst42 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+
+ ILVRL_H2_SH(dst31, dst20, dst10, dst32);
+ ILVRL_H2_SH(dst42, dst31, dst21, dst43);
+
+ dst0 = HEVC_FILT_4TAP(dst10, dst32, filt_h0, filt_h1);
+ dst1 = HEVC_FILT_4TAP(dst21, dst43, filt_h0, filt_h1);
+ dst0 >>= 6;
+ dst1 >>= 6;
+ tmp = __msa_pckev_h((v8i16) dst1, (v8i16) dst0);
+ tmp = __msa_srari_h(tmp, 6);
+ tmp = __msa_sat_s_h(tmp, 7);
+ out = PCKEV_XORI128_UB(tmp, tmp);
+ ST4x2_UB(out, dst, dst_stride);
}
static void hevc_hv_uni_4t_4x4_msa(uint8_t *src,
@@ -3190,20 +3145,19 @@ static void hevc_hv_uni_4t_4x4_msa(uint8_t *src,
uint8_t *dst,
int32_t dst_stride,
const int8_t *filter_x,
- const int8_t *filter_y,
- int32_t height)
+ const int8_t *filter_y)
{
+ v16u8 out;
v16i8 src0, src1, src2, src3, src4, src5, src6;
v8i16 filt0, filt1;
- v4i32 filt_h0, filt_h1;
- v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+ v8i16 filt_h0, filt_h1;
+ v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
v16i8 mask1;
- v8i16 filter_vec, const_vec;
- v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
- v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
- v4i32 dst0_r, dst1_r, dst2_r, dst3_r;
- v8i16 out0_r, out1_r;
- v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
+ v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8i16 filter_vec, tmp0, tmp1;
+ v8i16 dst30, dst41, dst52, dst63;
+ v8i16 dst10, dst32, dst54, dst21, dst43, dst65;
+ v4i32 dst0, dst1, dst2, dst3;
src -= (src_stride + 1);
@@ -3211,78 +3165,38 @@ static void hevc_hv_uni_4t_4x4_msa(uint8_t *src,
SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
filter_vec = LD_SH(filter_y);
- vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
- filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
- SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
mask1 = mask0 + 2;
- const_vec = __msa_ldi_h(128);
- const_vec <<= 6;
-
- LD_SB3(src, src_stride, src0, src1, src2);
- src += (3 * src_stride);
-
- XORI_B3_128_SB(src0, src1, src2);
-
- VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
- VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
- VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
-
- dst0 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
- dst1 = const_vec;
- DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
-
- ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r);
- LD_SB4(src, src_stride, src3, src4, src5, src6);
- XORI_B4_128_SB(src3, src4, src5, src6);
-
- /* row 3 */
- VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
- dst3 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
-
- dst32_r = __msa_ilvr_h(dst3, dst2);
- dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
- dst0_r >>= 6;
-
- /* row 4 */
- VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
- dst4 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
-
- dst43_r = __msa_ilvr_h(dst4, dst3);
- dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
- dst1_r >>= 6;
-
- /* row 5 */
- VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
- dst5 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
-
- dst10_r = __msa_ilvr_h(dst5, dst4);
- dst2_r = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1);
- dst2_r >>= 6;
-
- /* row 6 */
- VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
-
- dst21_r = __msa_ilvr_h(dst2, dst5);
- dst3_r = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1);
- dst3_r >>= 6;
-
- PCKEV_H2_SH(dst1_r, dst0_r, dst3_r, dst2_r, out0_r, out1_r);
- SRARI_H2_SH(out0_r, out1_r, 6);
- CLIP_SH2_0_255(out0_r, out1_r);
- out0_r = (v8i16) __msa_pckev_b((v16i8) out1_r, (v16i8) out0_r);
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- ST4x4_UB(out0_r, out0_r, 0, 1, 2, 3, dst, dst_stride);
+ VSHF_B2_SB(src0, src3, src0, src3, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src1, src4, src1, src4, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src2, src5, src2, src5, mask0, mask1, vec4, vec5);
+ VSHF_B2_SB(src3, src6, src3, src6, mask0, mask1, vec6, vec7);
+
+ dst30 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst41 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst52 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dst63 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+ ILVRL_H2_SH(dst41, dst30, dst10, dst43);
+ ILVRL_H2_SH(dst52, dst41, dst21, dst54);
+ ILVRL_H2_SH(dst63, dst52, dst32, dst65);
+ dst0 = HEVC_FILT_4TAP(dst10, dst32, filt_h0, filt_h1);
+ dst1 = HEVC_FILT_4TAP(dst21, dst43, filt_h0, filt_h1);
+ dst2 = HEVC_FILT_4TAP(dst32, dst54, filt_h0, filt_h1);
+ dst3 = HEVC_FILT_4TAP(dst43, dst65, filt_h0, filt_h1);
+ SRA_4V(dst0, dst1, dst2, dst3, 6);
+ PCKEV_H2_SH(dst1, dst0, dst3, dst2, tmp0, tmp1);
+ SRARI_H2_SH(tmp0, tmp1, 6);
+ SAT_SH2_SH(tmp0, tmp1, 7);
+ out = PCKEV_XORI128_UB(tmp0, tmp1);
+ ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
}
static void hevc_hv_uni_4t_4multx8mult_msa(uint8_t *src,
@@ -3294,19 +3208,20 @@ static void hevc_hv_uni_4t_4multx8mult_msa(uint8_t *src,
int32_t height)
{
uint32_t loop_cnt;
+ v16u8 out0, out1;
v16i8 src0, src1, src2, src3, src4, src5;
v16i8 src6, src7, src8, src9, src10;
v8i16 filt0, filt1;
- v4i32 filt_h0, filt_h1;
- v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+ v8i16 filt_h0, filt_h1;
+ v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
v16i8 mask1;
- v8i16 filter_vec, const_vec;
- v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
- v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8, dst9;
+ v8i16 filter_vec, tmp0, tmp1, tmp2, tmp3;
+ v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8i16 dst10, dst21, dst22, dst73, dst84, dst95, dst106;
v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r, dst6_r, dst7_r;
v8i16 dst10_r, dst32_r, dst54_r, dst76_r;
v8i16 dst21_r, dst43_r, dst65_r, dst87_r;
- v8i16 out0_r, out1_r, out2_r, out3_r;
+ v8i16 dst98_r, dst109_r;
src -= (src_stride + 1);
@@ -3314,33 +3229,23 @@ static void hevc_hv_uni_4t_4multx8mult_msa(uint8_t *src,
SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
filter_vec = LD_SH(filter_y);
- vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
- filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
- SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
mask1 = mask0 + 2;
- const_vec = __msa_ldi_h(128);
- const_vec <<= 6;
-
LD_SB3(src, src_stride, src0, src1, src2);
src += (3 * src_stride);
XORI_B3_128_SB(src0, src1, src2);
- VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
- VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
- VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
-
- dst0 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
- dst1 = const_vec;
- DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
-
- ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r);
+ VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src1, src2, src1, src2, mask0, mask1, vec2, vec3);
+ dst10 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst21 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ ILVRL_H2_SH(dst21, dst10, dst10_r, dst21_r);
+ dst22 = (v8i16) __msa_splati_d((v2i64) dst21, 1);
for (loop_cnt = height >> 3; loop_cnt--;) {
LD_SB8(src, src_stride,
@@ -3349,88 +3254,46 @@ static void hevc_hv_uni_4t_4multx8mult_msa(uint8_t *src,
XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
- /* row 3 */
- VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
- dst3 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+ VSHF_B2_SB(src3, src7, src3, src7, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src4, src8, src4, src8, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src5, src9, src5, src9, mask0, mask1, vec4, vec5);
+ VSHF_B2_SB(src6, src10, src6, src10, mask0, mask1, vec6, vec7);
- dst32_r = __msa_ilvr_h(dst3, dst2);
- dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
- dst0_r >>= 6;
+ dst73 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst84 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst95 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dst106 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
- /* row 4 */
- VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
- dst4 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
+ dst32_r = __msa_ilvr_h(dst73, dst22);
+ ILVRL_H2_SH(dst84, dst73, dst43_r, dst87_r);
+ ILVRL_H2_SH(dst95, dst84, dst54_r, dst98_r);
+ ILVRL_H2_SH(dst106, dst95, dst65_r, dst109_r);
+ dst22 = (v8i16) __msa_splati_d((v2i64) dst73, 1);
+ dst76_r = __msa_ilvr_h(dst22, dst106);
- dst43_r = __msa_ilvr_h(dst4, dst3);
+ dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
- dst1_r >>= 6;
-
- /* row 5 */
- VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
- dst5 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
-
- dst54_r = __msa_ilvr_h(dst5, dst4);
dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
- dst2_r >>= 6;
-
- /* row 6 */
- VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
- dst6 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6);
-
- dst65_r = __msa_ilvr_h(dst6, dst5);
dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
- dst3_r >>= 6;
-
- /* row 7 */
- VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
- dst7 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7);
-
- dst76_r = __msa_ilvr_h(dst7, dst6);
dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
- dst4_r >>= 6;
-
- /* row 8 */
- VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec0, vec1);
- dst8 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst8, dst8);
-
- dst87_r = __msa_ilvr_h(dst8, dst7);
dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
- dst5_r >>= 6;
-
- /* row 9 */
- VSHF_B2_SB(src9, src9, src9, src9, mask0, mask1, vec0, vec1);
- dst9 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst9, dst9);
-
- dst10_r = __msa_ilvr_h(dst9, dst8);
- dst6_r = HEVC_FILT_4TAP(dst76_r, dst10_r, filt_h0, filt_h1);
- dst6_r >>= 6;
-
- /* row 10 */
- VSHF_B2_SB(src10, src10, src10, src10, mask0, mask1, vec0, vec1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
-
- dst21_r = __msa_ilvr_h(dst2, dst9);
- dst7_r = HEVC_FILT_4TAP(dst87_r, dst21_r, filt_h0, filt_h1);
- dst7_r >>= 6;
-
+ dst6_r = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
+ dst7_r = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
+ SRA_4V(dst0_r, dst1_r, dst2_r, dst3_r, 6);
+ SRA_4V(dst4_r, dst5_r, dst6_r, dst7_r, 6);
PCKEV_H4_SH(dst1_r, dst0_r, dst3_r, dst2_r,
dst5_r, dst4_r, dst7_r, dst6_r,
- out0_r, out1_r, out2_r, out3_r);
-
- SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, 6);
- CLIP_SH4_0_255(out0_r, out1_r, out2_r, out3_r);
-
- PCKEV_B2_SH(out1_r, out0_r, out3_r, out2_r, out0_r, out1_r);
- ST4x8_UB(out0_r, out1_r, dst, dst_stride);
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 6);
+ SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+ out0 = PCKEV_XORI128_UB(tmp0, tmp1);
+ out1 = PCKEV_XORI128_UB(tmp2, tmp3);
+ ST4x8_UB(out0, out1, dst, dst_stride);
dst += (8 * dst_stride);
+
+ dst10_r = dst98_r;
+ dst21_r = dst109_r;
+ dst22 = (v8i16) __msa_splati_d((v2i64) dst106, 1);
}
}
@@ -3444,10 +3307,10 @@ static void hevc_hv_uni_4t_4w_msa(uint8_t *src,
{
if (2 == height) {
hevc_hv_uni_4t_4x2_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height);
+ filter_x, filter_y);
} else if (4 == height) {
hevc_hv_uni_4t_4x4_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height);
+ filter_x, filter_y);
} else if (0 == (height % 8)) {
hevc_hv_uni_4t_4multx8mult_msa(src, src_stride, dst, dst_stride,
filter_x, filter_y, height);
@@ -3462,19 +3325,23 @@ static void hevc_hv_uni_4t_6w_msa(uint8_t *src,
const int8_t *filter_y,
int32_t height)
{
- uint32_t loop_cnt;
+ v16u8 out0, out1, out2;
v16i8 src0, src1, src2, src3, src4, src5, src6;
+ v16i8 src7, src8, src9, src10;
v8i16 filt0, filt1;
- v4i32 filt_h0, filt_h1;
- v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+ v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
v16i8 mask1;
- v8i16 filter_vec, const_vec;
- v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
- v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
+ v8i16 filt_h0, filt_h1, filter_vec;
+ v8i16 dsth0, dsth1, dsth2, dsth3, dsth4, dsth5, dsth6, dsth7, dsth8, dsth9;
+ v8i16 dsth10, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
+ v4i32 dst4_r, dst5_r, dst6_r, dst7_r;
v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
v8i16 dst10_l, dst32_l, dst21_l, dst43_l;
- v8i16 out0_r, out1_r, out2_r, out3_r;
+ v8i16 dst54_r, dst76_r, dst98_r, dst65_r, dst87_r, dst109_r;
+ v8i16 dst98_l, dst65_l, dst54_l, dst76_l, dst87_l, dst109_l;
+ v8i16 dst1021_l, dst3243_l, dst5465_l, dst7687_l, dst98109_l;
src -= (src_stride + 1);
@@ -3482,16 +3349,12 @@ static void hevc_hv_uni_4t_6w_msa(uint8_t *src,
SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
filter_vec = LD_SH(filter_y);
- vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
- filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
- SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
mask1 = mask0 + 2;
- const_vec = __msa_ldi_h(128);
- const_vec <<= 6;
-
LD_SB3(src, src_stride, src0, src1, src2);
src += (3 * src_stride);
@@ -3501,79 +3364,78 @@ static void hevc_hv_uni_4t_6w_msa(uint8_t *src,
VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
- dst0 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
- dst1 = const_vec;
- DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
-
- ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
- ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
-
- for (loop_cnt = height >> 2; loop_cnt--;) {
- LD_SB4(src, src_stride, src3, src4, src5, src6);
- src += (4 * src_stride);
-
- XORI_B4_128_SB(src3, src4, src5, src6);
-
- /* row 3 */
- VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
- dst3 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
-
- ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
- dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
- dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
- dst0_r >>= 6;
- dst0_l >>= 6;
-
- /* row 4 */
- VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
- dst4 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
-
- ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
- dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
- dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
- dst1_r >>= 6;
- dst1_l >>= 6;
-
- /* row 5 */
- VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
- dst5 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
-
- ILVRL_H2_SH(dst5, dst4, dst10_r, dst10_l);
- dst2_r = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1);
- dst2_l = HEVC_FILT_4TAP(dst32_l, dst10_l, filt_h0, filt_h1);
-
- dst2_r >>= 6;
- dst2_l >>= 6;
+ dsth0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dsth1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dsth2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
- /* row 6 */
- VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+ ILVRL_H2_SH(dsth1, dsth0, dst10_r, dst10_l);
+ ILVRL_H2_SH(dsth2, dsth1, dst21_r, dst21_l);
- ILVRL_H2_SH(dst2, dst5, dst21_r, dst21_l);
- dst3_r = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1);
- dst3_l = HEVC_FILT_4TAP(dst43_l, dst21_l, filt_h0, filt_h1);
+ LD_SB8(src, src_stride, src3, src4, src5, src6, src7, src8, src9, src10);
+ XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
- dst3_r >>= 6;
- dst3_l >>= 6;
+ VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
+ VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
- PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r,
- dst2_l, dst2_r, dst3_l, dst3_r,
- out0_r, out1_r, out2_r, out3_r);
+ dsth3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dsth4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dsth5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dsth6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
- SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, 6);
- CLIP_SH4_0_255(out0_r, out1_r, out2_r, out3_r);
+ VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src9, src9, src9, src9, mask0, mask1, vec4, vec5);
+ VSHF_B2_SB(src10, src10, src10, src10, mask0, mask1, vec6, vec7);
+
+ dsth7 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dsth8 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dsth9 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dsth10 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+ ILVRL_H2_SH(dsth3, dsth2, dst32_r, dst32_l);
+ ILVRL_H2_SH(dsth4, dsth3, dst43_r, dst43_l);
+ ILVRL_H2_SH(dsth5, dsth4, dst54_r, dst54_l);
+ ILVRL_H2_SH(dsth6, dsth5, dst65_r, dst65_l);
+ ILVRL_H2_SH(dsth7, dsth6, dst76_r, dst76_l);
+ ILVRL_H2_SH(dsth8, dsth7, dst87_r, dst87_l);
+ ILVRL_H2_SH(dsth9, dsth8, dst98_r, dst98_l);
+ ILVRL_H2_SH(dsth10, dsth9, dst109_r, dst109_l);
+
+ PCKEV_D2_SH(dst21_l, dst10_l, dst43_l, dst32_l, dst1021_l, dst3243_l);
+ PCKEV_D2_SH(dst65_l, dst54_l, dst87_l, dst76_l, dst5465_l, dst7687_l);
+ dst98109_l = (v8i16) __msa_pckev_d((v2i64) dst109_l, (v2i64) dst98_l);
- PCKEV_B2_SH(out1_r, out0_r, out3_r, out2_r, out0_r, out1_r);
- ST6x4_UB(out0_r, out1_r, dst, dst_stride);
- dst += (4 * dst_stride);
- }
+ dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+ dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+ dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+ dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+ dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
+ dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
+ dst6_r = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
+ dst7_r = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
+ dst0_l = HEVC_FILT_4TAP(dst1021_l, dst3243_l, filt_h0, filt_h1);
+ dst1_l = HEVC_FILT_4TAP(dst3243_l, dst5465_l, filt_h0, filt_h1);
+ dst2_l = HEVC_FILT_4TAP(dst5465_l, dst7687_l, filt_h0, filt_h1);
+ dst3_l = HEVC_FILT_4TAP(dst7687_l, dst98109_l, filt_h0, filt_h1);
+ SRA_4V(dst0_r, dst1_r, dst2_r, dst3_r, 6);
+ SRA_4V(dst4_r, dst5_r, dst6_r, dst7_r, 6);
+ SRA_4V(dst0_l, dst1_l, dst2_l, dst3_l, 6);
+ PCKEV_H2_SH(dst1_r, dst0_r, dst3_r, dst2_r, tmp0, tmp1);
+ PCKEV_H2_SH(dst5_r, dst4_r, dst7_r, dst6_r, tmp2, tmp3);
+ PCKEV_H2_SH(dst1_l, dst0_l, dst3_l, dst2_l, tmp4, tmp5);
+ SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 6);
+ SRARI_H2_SH(tmp4, tmp5, 6);
+ SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3,7);
+ SAT_SH2_SH(tmp4, tmp5,7);
+ out0 = PCKEV_XORI128_UB(tmp0, tmp1);
+ out1 = PCKEV_XORI128_UB(tmp2, tmp3);
+ out2 = PCKEV_XORI128_UB(tmp4, tmp5);
+ ST4x8_UB(out0, out1, dst, dst_stride);
+ ST2x4_UB(out2, 0, dst + 4, dst_stride);
+ dst += 4 * dst_stride;
+ ST2x4_UB(out2, 4, dst + 4, dst_stride);
}
static void hevc_hv_uni_4t_8x2_msa(uint8_t *src,
@@ -3581,16 +3443,15 @@ static void hevc_hv_uni_4t_8x2_msa(uint8_t *src,
uint8_t *dst,
int32_t dst_stride,
const int8_t *filter_x,
- const int8_t *filter_y,
- int32_t height)
+ const int8_t *filter_y)
{
+ v16u8 out;
v16i8 src0, src1, src2, src3, src4;
v8i16 filt0, filt1;
- v4i32 filt_h0, filt_h1;
- v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+ v8i16 filt_h0, filt_h1, filter_vec;
+ v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
v16i8 mask1;
- v8i16 filter_vec, const_vec;
- v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+ v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9;
v8i16 dst0, dst1, dst2, dst3, dst4;
v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
@@ -3603,66 +3464,125 @@ static void hevc_hv_uni_4t_8x2_msa(uint8_t *src,
SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
filter_vec = LD_SH(filter_y);
- vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
- filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
- SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
mask1 = mask0 + 2;
- const_vec = __msa_ldi_h(128);
- const_vec <<= 6;
-
- LD_SB3(src, src_stride, src0, src1, src2);
- src += (3 * src_stride);
-
- XORI_B3_128_SB(src0, src1, src2);
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+ XORI_B5_128_SB(src0, src1, src2, src3, src4);
VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
-
- dst0 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
- dst1 = const_vec;
- DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
-
+ VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
+ VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec8, vec9);
+
+ dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+ dst4 = HEVC_FILT_4TAP_SH(vec8, vec9, filt0, filt1);
ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
-
- LD_SB2(src, src_stride, src3, src4);
- XORI_B2_128_SB(src3, src4);
-
- /* row 3 */
- VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
- dst3 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
-
ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+ ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
- dst0_r >>= 6;
- dst0_l >>= 6;
-
- /* row 4 */
- VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
- dst4 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
-
- ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
- dst1_r >>= 6;
- dst1_l >>= 6;
-
+ SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
PCKEV_H2_SH(dst0_l, dst0_r, dst1_l, dst1_r, out0_r, out1_r);
SRARI_H2_SH(out0_r, out1_r, 6);
- CLIP_SH2_0_255(out0_r, out1_r);
- out0_r = (v8i16) __msa_pckev_b((v16i8) out1_r, (v16i8) out0_r);
+ SAT_SH2_SH(out0_r, out1_r, 7);
+ out = PCKEV_XORI128_UB(out0_r, out1_r);
+ ST8x2_UB(out, dst, dst_stride);
+}
+
+static void hevc_hv_uni_4t_8multx4_msa(uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int8_t *filter_x,
+ const int8_t *filter_y,
+ int32_t width8mult)
+{
+ uint32_t cnt;
+ v16u8 out0, out1;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, mask0, mask1;
+ v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8i16 filt0, filt1, filt_h0, filt_h1, filter_vec;
+ v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, tmp0, tmp1, tmp2, tmp3;
+ v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
+ v8i16 dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
+ v8i16 dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
- ST8x2_UB(out0_r, dst, dst_stride);
+ src -= (src_stride + 1);
+
+ filter_vec = LD_SH(filter_x);
+ SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+ filter_vec = LD_SH(filter_y);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
+
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
+
+ mask0 = LD_SB(ff_hevc_mask_arr);
+ mask1 = mask0 + 2;
+
+ for (cnt = width8mult; cnt--;) {
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ src += 8;
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+
+ VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+
+ dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+
+ ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
+ ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
+
+ VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
+ VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
+
+ dst3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dst6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+ ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+ ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+ ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
+ ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
+
+ dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+ dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
+ dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+ dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
+ dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+ dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
+ dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+ dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
+
+ SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
+ SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
+
+ PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l,
+ dst3_r, tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 6);
+ SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+ out0 = PCKEV_XORI128_UB(tmp0, tmp1);
+ out1 = PCKEV_XORI128_UB(tmp2, tmp3);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += 8;
+ }
}
static void hevc_hv_uni_4t_8x6_msa(uint8_t *src,
@@ -3670,16 +3590,16 @@ static void hevc_hv_uni_4t_8x6_msa(uint8_t *src,
uint8_t *dst,
int32_t dst_stride,
const int8_t *filter_x,
- const int8_t *filter_y,
- int32_t height)
+ const int8_t *filter_y)
{
+ v16u8 out0, out1, out2;
v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
v8i16 filt0, filt1;
- v4i32 filt_h0, filt_h1;
- v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+ v8i16 filt_h0, filt_h1, filter_vec;
+ v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
v16i8 mask1;
- v8i16 filter_vec, const_vec;
- v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+ v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9;
+ v16i8 vec10, vec11, vec12, vec13, vec14, vec15, vec16, vec17;
v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
v4i32 dst4_r, dst4_l, dst5_r, dst5_l;
@@ -3695,157 +3615,104 @@ static void hevc_hv_uni_4t_8x6_msa(uint8_t *src,
SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
filter_vec = LD_SH(filter_y);
- vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
- filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
- SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
mask1 = mask0 + 2;
- const_vec = __msa_ldi_h(128);
- const_vec <<= 6;
-
- LD_SB3(src, src_stride, src0, src1, src2);
- src += (3 * src_stride);
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+ src += (5 * src_stride);
+ LD_SB4(src, src_stride, src5, src6, src7, src8);
- XORI_B3_128_SB(src0, src1, src2);
+ XORI_B5_128_SB(src0, src1, src2, src3, src4);
+ XORI_B4_128_SB(src5, src6, src7, src8);
VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
-
- dst0 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
- dst1 = const_vec;
- DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+ VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
+ VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec8, vec9);
+ VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec10, vec11);
+ VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec12, vec13);
+ VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec14, vec15);
+ VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec16, vec17);
+
+ dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+ dst4 = HEVC_FILT_4TAP_SH(vec8, vec9, filt0, filt1);
+ dst5 = HEVC_FILT_4TAP_SH(vec10, vec11, filt0, filt1);
+ dst6 = HEVC_FILT_4TAP_SH(vec12, vec13, filt0, filt1);
+ dst7 = HEVC_FILT_4TAP_SH(vec14, vec15, filt0, filt1);
+ dst8 = HEVC_FILT_4TAP_SH(vec16, vec17, filt0, filt1);
ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
-
- LD_SB2(src, src_stride, src3, src4);
- src += (2 * src_stride);
-
- XORI_B2_128_SB(src3, src4);
-
- /* row 3 */
- VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
- dst3 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
-
ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+ ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+ ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
+ ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
+ ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l);
+ ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_l);
+
dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
-
- dst0_r >>= 6;
- dst0_l >>= 6;
-
- /* row 4 */
- VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
- dst4 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
-
- ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
- dst1_r >>= 6;
- dst1_l >>= 6;
-
- LD_SB2(src, src_stride, src5, src6);
- src += (2 * src_stride);
-
- XORI_B2_128_SB(src5, src6);
-
- /* row 5 */
- VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
- dst5 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
-
- ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
- dst2_r >>= 6;
- dst2_l >>= 6;
-
- /* row 6 */
- VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
- dst6 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6);
-
- ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
- dst3_r >>= 6;
- dst3_l >>= 6;
-
- LD_SB2(src, src_stride, src7, src8);
- src += (2 * src_stride);
-
- XORI_B2_128_SB(src7, src8);
-
- /* row 7 */
- VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
- dst7 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7);
-
- ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l);
dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
dst4_l = HEVC_FILT_4TAP(dst54_l, dst76_l, filt_h0, filt_h1);
-
- dst4_r >>= 6;
- dst4_l >>= 6;
-
- /* row 8 */
- VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec0, vec1);
- dst8 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst8, dst8);
-
- ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_l);
dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
dst5_l = HEVC_FILT_4TAP(dst65_l, dst87_l, filt_h0, filt_h1);
- dst5_r >>= 6;
- dst5_l >>= 6;
+ SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
+ SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
+ SRA_4V(dst4_r, dst4_l, dst5_r, dst5_l, 6);
PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r,
dst2_l, dst2_r, dst3_l, dst3_r, out0_r, out1_r, out2_r, out3_r);
PCKEV_H2_SH(dst4_l, dst4_r, dst5_l, dst5_r, out4_r, out5_r);
SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, 6);
SRARI_H2_SH(out4_r, out5_r, 6);
- CLIP_SH4_0_255(out0_r, out1_r, out2_r, out3_r);
- CLIP_SH2_0_255(out4_r, out5_r);
-
- PCKEV_B2_SH(out1_r, out0_r, out3_r, out2_r, out0_r, out1_r);
- out2_r = (v8i16) __msa_pckev_b((v16i8) out5_r, (v16i8) out4_r);
+ SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
+ SAT_SH2_SH(out4_r, out5_r, 7);
+ out0 = PCKEV_XORI128_UB(out0_r, out1_r);
+ out1 = PCKEV_XORI128_UB(out2_r, out3_r);
+ out2 = PCKEV_XORI128_UB(out4_r, out5_r);
- ST8x4_UB(out0_r, out1_r, dst, dst_stride);
+ ST8x4_UB(out0, out1, dst, dst_stride);
dst += (4 * dst_stride);
- ST8x2_UB(out2_r, dst, dst_stride);
+ ST8x2_UB(out2, dst, dst_stride);
}
-static void hevc_hv_uni_4t_8w_mult_msa(uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- const int8_t *filter_x,
- const int8_t *filter_y,
- int32_t height,
- int32_t width)
+static void hevc_hv_uni_4t_8multx4mult_msa(uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int8_t *filter_x,
+ const int8_t *filter_y,
+ int32_t height,
+ int32_t width8mult)
{
uint32_t loop_cnt, cnt;
uint8_t *src_tmp;
uint8_t *dst_tmp;
+ v16u8 out0, out1;
v16i8 src0, src1, src2, src3, src4, src5, src6;
v8i16 filt0, filt1;
- v4i32 filt_h0, filt_h1;
- v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 };
+ v8i16 filt_h0, filt_h1, filter_vec;
+ v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
v16i8 mask1;
- v8i16 filter_vec, const_vec;
- v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
+ v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
v8i16 dst10_l, dst32_l, dst21_l, dst43_l;
+ v8i16 dst54_r, dst54_l, dst65_r, dst65_l, dst6;
v8i16 out0_r, out1_r, out2_r, out3_r;
src -= (src_stride + 1);
@@ -3854,17 +3721,13 @@ static void hevc_hv_uni_4t_8w_mult_msa(uint8_t *src,
SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
filter_vec = LD_SH(filter_y);
- vec0 = __msa_clti_s_b((v16i8) filter_vec, 0);
- filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
- SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1);
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
mask1 = mask0 + 2;
- const_vec = __msa_ldi_h(128);
- const_vec <<= 6;
-
- for (cnt = width >> 3; cnt--;) {
+ for (cnt = width8mult; cnt--;) {
src_tmp = src;
dst_tmp = dst;
@@ -3877,79 +3740,62 @@ static void hevc_hv_uni_4t_8w_mult_msa(uint8_t *src,
VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
- dst0 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0);
- dst1 = const_vec;
- DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2);
+ dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
- for (loop_cnt = height >> 2; loop_cnt--;) {
+ for (loop_cnt = (height >> 2); loop_cnt--;) {
LD_SB4(src_tmp, src_stride, src3, src4, src5, src6);
src_tmp += (4 * src_stride);
XORI_B4_128_SB(src3, src4, src5, src6);
- /* row 3 */
VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
- dst3 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3);
+ VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
+ VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
+
+ dst3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dst6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
+ ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
+ ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
+ ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
+
dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
-
- dst0_r >>= 6;
- dst0_l >>= 6;
-
- /* row 4 */
- VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
- dst4 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4);
-
- ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
- dst1_r >>= 6;
- dst1_l >>= 6;
-
- /* row 5 */
- VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1);
- dst5 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5);
-
- ILVRL_H2_SH(dst5, dst4, dst10_r, dst10_l);
- dst2_r = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1);
- dst2_l = HEVC_FILT_4TAP(dst32_l, dst10_l, filt_h0, filt_h1);
-
- dst2_r >>= 6;
- dst2_l >>= 6;
-
- /* row 6 */
- VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1);
- dst2 = const_vec;
- DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2);
+ dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+ dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
+ dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+ dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
- ILVRL_H2_SH(dst2, dst5, dst21_r, dst21_l);
- dst3_r = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1);
- dst3_l = HEVC_FILT_4TAP(dst43_l, dst21_l, filt_h0, filt_h1);
-
- dst3_r >>= 6;
- dst3_l >>= 6;
+ SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
+ SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r,
dst2_l, dst2_r, dst3_l, dst3_r,
out0_r, out1_r, out2_r, out3_r);
SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, 6);
- CLIP_SH4_0_255(out0_r, out1_r, out2_r, out3_r);
-
- PCKEV_B2_SH(out1_r, out0_r, out3_r, out2_r, out0_r, out1_r);
- ST8x4_UB(out0_r, out1_r, dst_tmp, dst_stride);
+ SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
+ out0 = PCKEV_XORI128_UB(out0_r, out1_r);
+ out1 = PCKEV_XORI128_UB(out2_r, out3_r);
+ ST8x4_UB(out0, out1, dst_tmp, dst_stride);
dst_tmp += (4 * dst_stride);
+
+ dst10_r = dst54_r;
+ dst10_l = dst54_l;
+ dst21_r = dst65_r;
+ dst21_l = dst65_l;
+ dst2 = dst6;
}
src += 8;
@@ -3967,13 +3813,16 @@ static void hevc_hv_uni_4t_8w_msa(uint8_t *src,
{
if (2 == height) {
hevc_hv_uni_4t_8x2_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height);
+ filter_x, filter_y);
+ } else if (4 == height) {
+ hevc_hv_uni_4t_8multx4_msa(src, src_stride, dst, dst_stride,
+ filter_x, filter_y, 1);
} else if (6 == height) {
hevc_hv_uni_4t_8x6_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height);
+ filter_x, filter_y);
} else if (0 == (height % 4)) {
- hevc_hv_uni_4t_8w_mult_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height, 8);
+ hevc_hv_uni_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+ filter_x, filter_y, height, 1);
}
}
@@ -3985,11 +3834,164 @@ static void hevc_hv_uni_4t_12w_msa(uint8_t *src,
const int8_t *filter_y,
int32_t height)
{
- hevc_hv_uni_4t_8w_mult_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height, 8);
+ uint32_t loop_cnt;
+ uint8_t *src_tmp, *dst_tmp;
+ v16u8 out0, out1;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v16i8 mask0, mask1, mask2, mask3;
+ v8i16 filt0, filt1, filt_h0, filt_h1, filter_vec, tmp0, tmp1, tmp2, tmp3;
+ v8i16 dsth0, dsth1, dsth2, dsth3, dsth4, dsth5, dsth6;
+ v8i16 dst10, dst21, dst22, dst73, dst84, dst95, dst106;
+ v8i16 dst76_r, dst98_r, dst87_r, dst109_r;
+ v8i16 dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
+ v8i16 dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
+ v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
+ v4i32 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- hevc_hv_uni_4t_4w_msa(src + 8, src_stride, dst + 8, dst_stride,
- filter_x, filter_y, height);
+ src -= (src_stride + 1);
+
+ filter_vec = LD_SH(filter_x);
+ SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
+
+ filter_vec = LD_SH(filter_y);
+ UNPCK_R_SB_SH(filter_vec, filter_vec);
+
+ SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
+
+ mask0 = LD_SB(ff_hevc_mask_arr);
+ mask1 = mask0 + 2;
+
+ src_tmp = src;
+ dst_tmp = dst;
+
+ LD_SB3(src_tmp, src_stride, src0, src1, src2);
+ src_tmp += (3 * src_stride);
+
+ XORI_B3_128_SB(src0, src1, src2);
+
+ VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
+
+ dsth0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dsth1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dsth2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+
+ ILVRL_H2_SH(dsth1, dsth0, dst10_r, dst10_l);
+ ILVRL_H2_SH(dsth2, dsth1, dst21_r, dst21_l);
+
+ for (loop_cnt = 4; loop_cnt--;) {
+ LD_SB4(src_tmp, src_stride, src3, src4, src5, src6);
+ src_tmp += (4 * src_stride);
+ XORI_B4_128_SB(src3, src4, src5, src6);
+
+ VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
+ VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
+ VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
+ VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
+
+ dsth3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dsth4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dsth5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dsth6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+ ILVRL_H2_SH(dsth3, dsth2, dst32_r, dst32_l);
+ ILVRL_H2_SH(dsth4, dsth3, dst43_r, dst43_l);
+ ILVRL_H2_SH(dsth5, dsth4, dst54_r, dst54_l);
+ ILVRL_H2_SH(dsth6, dsth5, dst65_r, dst65_l);
+
+ dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+ dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
+ dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+ dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
+ dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+ dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
+ dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+ dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
+
+ SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
+ SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
+
+ PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l,
+ dst3_r, tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 6);
+ SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+ out0 = PCKEV_XORI128_UB(tmp0, tmp1);
+ out1 = PCKEV_XORI128_UB(tmp2, tmp3);
+ ST8x4_UB(out0, out1, dst_tmp, dst_stride);
+ dst_tmp += (4 * dst_stride);
+
+ dst10_r = dst54_r;
+ dst10_l = dst54_l;
+ dst21_r = dst65_r;
+ dst21_l = dst65_l;
+ dsth2 = dsth6;
+ }
+
+ src += 8;
+ dst += 8;
+
+ mask2 = LD_SB(ff_hevc_mask_arr + 16);
+ mask3 = mask2 + 2;
+
+ LD_SB3(src, src_stride, src0, src1, src2);
+ src += (3 * src_stride);
+ XORI_B3_128_SB(src0, src1, src2);
+ VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1);
+ VSHF_B2_SB(src1, src2, src1, src2, mask2, mask3, vec2, vec3);
+
+ dst10 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst21 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+
+ ILVRL_H2_SH(dst21, dst10, dst10_r, dst21_r);
+ dst22 = (v8i16) __msa_splati_d((v2i64) dst21, 1);
+
+ for (loop_cnt = 2; loop_cnt--;) {
+ LD_SB8(src, src_stride,
+ src3, src4, src5, src6, src7, src8, src9, src10);
+ src += (8 * src_stride);
+ XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
+ VSHF_B2_SB(src3, src7, src3, src7, mask2, mask3, vec0, vec1);
+ VSHF_B2_SB(src4, src8, src4, src8, mask2, mask3, vec2, vec3);
+ VSHF_B2_SB(src5, src9, src5, src9, mask2, mask3, vec4, vec5);
+ VSHF_B2_SB(src6, src10, src6, src10, mask2, mask3, vec6, vec7);
+
+ dst73 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
+ dst84 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
+ dst95 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
+ dst106 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
+
+ dst32_r = __msa_ilvr_h(dst73, dst22);
+ ILVRL_H2_SH(dst84, dst73, dst43_r, dst87_r);
+ ILVRL_H2_SH(dst95, dst84, dst54_r, dst98_r);
+ ILVRL_H2_SH(dst106, dst95, dst65_r, dst109_r);
+ dst22 = (v8i16) __msa_splati_d((v2i64) dst73, 1);
+ dst76_r = __msa_ilvr_h(dst22, dst106);
+
+ dst0 = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
+ dst1 = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
+ dst2 = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
+ dst3 = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
+ dst4 = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
+ dst5 = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
+ dst6 = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
+ dst7 = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
+ SRA_4V(dst0, dst1, dst2, dst3, 6);
+ SRA_4V(dst4, dst5, dst6, dst7, 6);
+ PCKEV_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6,
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, 6);
+ SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+ out0 = PCKEV_XORI128_UB(tmp0, tmp1);
+ out1 = PCKEV_XORI128_UB(tmp2, tmp3);
+ ST4x8_UB(out0, out1, dst, dst_stride);
+ dst += (8 * dst_stride);
+
+ dst10_r = dst98_r;
+ dst21_r = dst109_r;
+ dst22 = (v8i16) __msa_splati_d((v2i64) dst106, 1);
+ }
}
static void hevc_hv_uni_4t_16w_msa(uint8_t *src,
@@ -4000,8 +4002,13 @@ static void hevc_hv_uni_4t_16w_msa(uint8_t *src,
const int8_t *filter_y,
int32_t height)
{
- hevc_hv_uni_4t_8w_mult_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height, 16);
+ if (4 == height) {
+ hevc_hv_uni_4t_8multx4_msa(src, src_stride, dst, dst_stride, filter_x,
+ filter_y, 2);
+ } else {
+ hevc_hv_uni_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+ filter_x, filter_y, height, 2);
+ }
}
static void hevc_hv_uni_4t_24w_msa(uint8_t *src,
@@ -4012,8 +4019,8 @@ static void hevc_hv_uni_4t_24w_msa(uint8_t *src,
const int8_t *filter_y,
int32_t height)
{
- hevc_hv_uni_4t_8w_mult_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height, 24);
+ hevc_hv_uni_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+ filter_x, filter_y, height, 3);
}
static void hevc_hv_uni_4t_32w_msa(uint8_t *src,
@@ -4024,8 +4031,8 @@ static void hevc_hv_uni_4t_32w_msa(uint8_t *src,
const int8_t *filter_y,
int32_t height)
{
- hevc_hv_uni_4t_8w_mult_msa(src, src_stride, dst, dst_stride,
- filter_x, filter_y, height, 32);
+ hevc_hv_uni_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
+ filter_x, filter_y, height, 4);
}
#define UNI_MC_COPY(WIDTH) \
--
1.7.9.5
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel at ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
More information about the ffmpeg-devel
mailing list