[FFmpeg-devel] [PATCH] vp9: add avx2 iadst16 implementations.
Ronald S. Bultje
rsbultje at gmail.com
Tue Nov 8 18:22:31 EET 2016
Also a small cosmetic change to the avx2 idct16 version to make it
explicit that one of the arguments to the write-out macros is unused
for >=avx2 (it uses pmovzxbw instead of punpcklbw).
---
libavcodec/x86/vp9dsp_init.c | 8 +-
libavcodec/x86/vp9itxfm.asm | 188 ++++++++++++++++++++++++++++++++++++++++---
2 files changed, 183 insertions(+), 13 deletions(-)
diff --git a/libavcodec/x86/vp9dsp_init.c b/libavcodec/x86/vp9dsp_init.c
index cc781a0..66363d4 100644
--- a/libavcodec/x86/vp9dsp_init.c
+++ b/libavcodec/x86/vp9dsp_init.c
@@ -114,7 +114,7 @@ itxfm_func(idct, idct, 32, sse2);
itxfm_func(idct, idct, 32, ssse3);
itxfm_func(idct, idct, 32, avx);
itxfm_func(iwht, iwht, 4, mmx);
-itxfm_func(idct, idct, 16, avx2);
+itxfm_funcs(16, avx2);
itxfm_func(idct, idct, 32, avx2);
#undef itxfm_func
@@ -391,6 +391,12 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
if (ARCH_X86_64) {
#if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx2;
+ dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx2;
+ dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx2;
+ dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx2;
+ dsp->itxfm_add[TX_32X32][ADST_ADST] =
+ dsp->itxfm_add[TX_32X32][ADST_DCT] =
+ dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx2;
init_subpel3_32_64(0, put, 8, avx2);
init_subpel3_32_64(1, avg, 8, avx2);
diff --git a/libavcodec/x86/vp9itxfm.asm b/libavcodec/x86/vp9itxfm.asm
index 57d6d35..99c4f9d 100644
--- a/libavcodec/x86/vp9itxfm.asm
+++ b/libavcodec/x86/vp9itxfm.asm
@@ -1581,33 +1581,30 @@ cglobal vp9_idct_idct_16x16_add, 4, 4, 16, dst, stride, block, eob
VP9_IDCT16_YMM_1D
mova [blockq+224], m7
- mova [blockq+480], m15
- pxor m15, m15
; store
- VP9_IDCT8_WRITEx2 0, 1, 6, 7, 15, [pw_512], 6
+ VP9_IDCT8_WRITEx2 0, 1, 6, 7, unused, [pw_512], 6
lea dstq, [dstq+2*strideq]
- VP9_IDCT8_WRITEx2 2, 3, 6, 7, 15, [pw_512], 6
+ VP9_IDCT8_WRITEx2 2, 3, 6, 7, unused, [pw_512], 6
lea dstq, [dstq+2*strideq]
- VP9_IDCT8_WRITEx2 4, 5, 6, 7, 15, [pw_512], 6
+ VP9_IDCT8_WRITEx2 4, 5, 6, 7, unused, [pw_512], 6
lea dstq, [dstq+2*strideq]
mova m6, [blockq+192]
mova m7, [blockq+224]
- SWAP 0, 15
- mova m15, [blockq+480]
- VP9_IDCT8_WRITEx2 6, 7, 1, 2, 0, [pw_512], 6
+ VP9_IDCT8_WRITEx2 6, 7, 1, 2, unused, [pw_512], 6
lea dstq, [dstq+2*strideq]
- VP9_IDCT8_WRITEx2 8, 9, 1, 2, 0, [pw_512], 6
+ VP9_IDCT8_WRITEx2 8, 9, 1, 2, unused, [pw_512], 6
lea dstq, [dstq+2*strideq]
- VP9_IDCT8_WRITEx2 10, 11, 1, 2, 0, [pw_512], 6
+ VP9_IDCT8_WRITEx2 10, 11, 1, 2, unused, [pw_512], 6
lea dstq, [dstq+2*strideq]
- VP9_IDCT8_WRITEx2 12, 13, 1, 2, 0, [pw_512], 6
+ VP9_IDCT8_WRITEx2 12, 13, 1, 2, unused, [pw_512], 6
lea dstq, [dstq+2*strideq]
- VP9_IDCT8_WRITEx2 14, 15, 1, 2, 0, [pw_512], 6
+ VP9_IDCT8_WRITEx2 14, 15, 1, 2, unused, [pw_512], 6
lea dstq, [dstq+2*strideq]
; at the end of the loop, m0 should still be zero
; use that to zero out block coefficients
+ pxor m0, m0
ZERO_BLOCK blockq, 32, 16, m0
RET
%endif
@@ -1987,6 +1984,173 @@ IADST16_FN idct, IDCT16, iadst, IADST16, avx
IADST16_FN iadst, IADST16, idct, IDCT16, avx
IADST16_FN iadst, IADST16, iadst, IADST16, avx
+; in: data in m[0-15] except m0/m4, which are in [blockq+0] and [blockq+128]
+; out: m[0-15] except m6, which is in [blockq+192]
+; uses blockq as scratch space
+%macro VP9_IADST16_YMM_1D 0
+ mova [blockq+ 32], m3
+ mova [blockq+ 64], m7
+ mova [blockq+ 96], m8
+
+ ; first half of round 1
+ VP9_UNPACK_MULSUB_2D_4X 9, 6, 0, 3, 13160, 9760 ; m9/x=t7[d], m6/x=t6[d]
+ VP9_UNPACK_MULSUB_2D_4X 1, 14, 4, 7, 2404, 16207 ; m1/x=t15[d], m14/x=t14[d]
+ VP9_RND_SH_SUMSUB_BA 14, 6, 7, 3, 8, [pd_8192] ; m14=t6[w], m6=t14[w]
+ VP9_RND_SH_SUMSUB_BA 1, 9, 4, 0, 8, [pd_8192] ; m1=t7[w], m9=t15[w]
+
+ VP9_UNPACK_MULSUB_2D_4X 13, 2, 4, 7, 15893, 3981 ; m13/x=t3[d], m2/x=t2[d]
+ VP9_UNPACK_MULSUB_2D_4X 5, 10, 0, 3, 8423, 14053 ; m5/x=t11[d], m10/x=t10[d]
+ VP9_RND_SH_SUMSUB_BA 10, 2, 3, 7, 8, [pd_8192] ; m10=t2[w], m2=t10[w]
+ VP9_RND_SH_SUMSUB_BA 5, 13, 0, 4, 8, [pd_8192] ; m5=t3[w], m13=t11[w]
+
+ ; half of round 2 t8-15
+ VP9_UNPACK_MULSUB_2D_4X 2, 13, 4, 7, 9102, 13623 ; m2/x=t11[d], m13/x=t10[d]
+ VP9_UNPACK_MULSUB_2D_4X 9, 6, 3, 0, 13623, 9102 ; m9/x=t14[d], m6/x=t15[d]
+ VP9_RND_SH_SUMSUB_BA 9, 13, 3, 7, 8, [pd_8192] ; m9=t10[w], m13=t14[w]
+ VP9_RND_SH_SUMSUB_BA 6, 2, 0, 4, 8, [pd_8192] ; m6=t11[w], m2=t15[w]
+
+ SUMSUB_BA w, 14, 10, 8 ; m14=t2, m10=t6
+ SUMSUB_BA w, 1, 5, 8 ; m1=t3, m5=t7
+
+ mova m0, [blockq+ 0]
+ mova m4, [blockq+128]
+ mova m3, [blockq+ 32]
+ mova m7, [blockq+ 64]
+ mova m8, [blockq+ 96]
+ mova [blockq+ 0], m1
+ mova [blockq+128], m14
+ mova [blockq+ 32], m6
+ mova [blockq+ 64], m9
+ mova [blockq+ 96], m10
+
+ ; second half of round 1
+ VP9_UNPACK_MULSUB_2D_4X 15, 0, 1, 9, 16364, 804 ; m15/x=t1[d], m0/x=t0[d]
+ VP9_UNPACK_MULSUB_2D_4X 7, 8, 10, 6, 11003, 12140 ; m7/x=t9[d], m8/x=t8[d]
+ VP9_RND_SH_SUMSUB_BA 8, 0, 6, 9, 14, [pd_8192] ; m8=t0[w], m0=t8[w]
+ VP9_RND_SH_SUMSUB_BA 7, 15, 10, 1, 14, [pd_8192] ; m7=t1[w], m15=t9[w]
+
+ VP9_UNPACK_MULSUB_2D_4X 11, 4, 10, 6, 14811, 7005 ; m11/x=t5[d], m4/x=t4[d]
+ VP9_UNPACK_MULSUB_2D_4X 3, 12, 1, 9, 5520, 15426 ; m3/x=t13[d], m12/x=t12[d]
+ VP9_RND_SH_SUMSUB_BA 12, 4, 9, 6, 14, [pd_8192] ; m12=t4[w], m4=t12[w]
+ VP9_RND_SH_SUMSUB_BA 3, 11, 1, 10, 14, [pd_8192] ; m3=t5[w], m11=t13[w]
+
+ ; second half of round 2 t8-15
+ VP9_UNPACK_MULSUB_2D_4X 0, 15, 6, 10, 16069, 3196 ; m15/x=t8[d], m0/x=t9[d]
+ VP9_UNPACK_MULSUB_2D_4X 11, 4, 9, 1, 3196, 16069 ; m11/x=t12[d], m4/x=t13[d]
+ VP9_RND_SH_SUMSUB_BA 11, 15, 9, 10, 14, [pd_8192] ; m11=t8[w], m15=t12[w]
+ VP9_RND_SH_SUMSUB_BA 4, 0, 1, 6, 14, [pd_8192] ; m4=t9[w], m0=t13[w]
+
+ SUMSUB_BA w, 12, 8, 14 ; m12=t0, m8=t4
+ SUMSUB_BA w, 3, 7, 14 ; m3=t1, m7=t5
+
+ mova m10, [blockq+ 96]
+ mova [blockq+ 96], m12
+
+ ; round 3
+ VP9_UNPACK_MULSUB_2D_4X 15, 0, 9, 12, 15137, 6270 ; m15/x=t13[d], m0/x=t12[d]
+ VP9_UNPACK_MULSUB_2D_4X 2, 13, 1, 6, 6270, 15137 ; m2/x=t14[d], m13/x=t15[d]
+ VP9_RND_SH_SUMSUB_BA 2, 0, 1, 12, 14, [pd_8192] ; m2=out2[w], m0=t14a[w]
+ VP9_RND_SH_SUMSUB_BA 13, 15, 6, 9, 14, [pd_8192]
+ PSIGNW m13, [pw_m1] ; m13=out13[w], m15=t15a[w]
+
+ VP9_UNPACK_MULSUB_2D_4X 8, 7, 12, 9, 15137, 6270 ; m8/x=t5[d], m7/x=t4[d]
+ VP9_UNPACK_MULSUB_2D_4X 5, 10, 1, 6, 6270, 15137 ; m5/x=t6[d], m10/x=t7[d]
+ VP9_RND_SH_SUMSUB_BA 5, 7, 1, 9, 14, [pd_8192]
+ PSIGNW m5, [pw_m1] ; m5=out3[w], m7=t6[w]
+ VP9_RND_SH_SUMSUB_BA 10, 8, 6, 12, 14, [pd_8192] ; m10=out12[w], m8=t7[w]
+
+ mova m1, [blockq+ 0]
+ mova m14, [blockq+128]
+ mova m6, [blockq+ 32]
+ mova m9, [blockq+ 64]
+ mova m12, [blockq+ 96]
+ mova [blockq+ 0], m10
+ mova [blockq+128], m5
+
+ SUMSUB_BA w, 14, 12, 5 ; m14=out0, m12=t2a
+ SUMSUB_BA w, 1, 3, 5
+ PSIGNW m1, [pw_m1] ; m1=out15, m3=t3a
+
+ SUMSUB_BA w, 9, 11, 5
+ PSIGNW m9, [pw_m1] ; m9=out1, m11=t10
+ SUMSUB_BA w, 6, 4, 5 ; m6=out14, m4=t11
+
+ VP9_UNPACK_MULSUB_2W_4X 4, 11, 11585, 11585, [pd_8192], 5, 10 ; m4=out9, m11=out6
+ mova m5, [blockq+128]
+ mova [blockq+192], m11
+ PSIGNW m15, [pw_m1]
+ VP9_UNPACK_MULSUB_2W_4X 15, 0, 11585, 11585, [pd_8192], 10, 11 ; m15=out5, m0=out10
+
+ PSIGNW m3, [pw_m1]
+ VP9_UNPACK_MULSUB_2W_4X 3, 12, 11585, 11585, [pd_8192], 10, 11 ; m3=out7,m12=out8
+ VP9_UNPACK_MULSUB_2W_4X 8, 7, 11585, 11585, [pd_8192], 10, 11 ; m8=out11,m7=out4
+
+ mova m10, [blockq+ 0]
+
+ SWAP 0, 14, 6, 11, 8, 12, 10
+ SWAP 1, 9, 15, 4, 7, 3, 5
+ SWAP 5, 9, 15
+%endmacro
+
+%if ARCH_X86_64
+%macro IADST16_YMM_FN 4
+INIT_YMM avx2
+cglobal vp9_%1_%3_16x16_add, 4, 4, 16, dst, stride, block, eob
+ mova m1, [blockq+ 32]
+ mova m2, [blockq+ 64]
+ mova m3, [blockq+ 96]
+ mova m5, [blockq+160]
+ mova m6, [blockq+192]
+ mova m7, [blockq+224]
+ mova m8, [blockq+256]
+ mova m9, [blockq+288]
+ mova m10, [blockq+320]
+ mova m11, [blockq+352]
+ mova m12, [blockq+384]
+ mova m13, [blockq+416]
+ mova m14, [blockq+448]
+ mova m15, [blockq+480]
+
+ VP9_%2_YMM_1D
+ TRANSPOSE16x16W 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, \
+ [blockq+192], [blockq+128], 1
+ mova [blockq+ 0], m0
+ VP9_%4_YMM_1D
+
+ mova [blockq+224], m7
+
+ ; store
+ VP9_IDCT8_WRITEx2 0, 1, 6, 7, unused, [pw_512], 6
+ lea dstq, [dstq+2*strideq]
+ VP9_IDCT8_WRITEx2 2, 3, 6, 7, unused, [pw_512], 6
+ lea dstq, [dstq+2*strideq]
+ VP9_IDCT8_WRITEx2 4, 5, 6, 7, unused, [pw_512], 6
+ lea dstq, [dstq+2*strideq]
+ mova m6, [blockq+192]
+ mova m7, [blockq+224]
+ VP9_IDCT8_WRITEx2 6, 7, 1, 2, unused, [pw_512], 6
+ lea dstq, [dstq+2*strideq]
+ VP9_IDCT8_WRITEx2 8, 9, 1, 2, unused, [pw_512], 6
+ lea dstq, [dstq+2*strideq]
+ VP9_IDCT8_WRITEx2 10, 11, 1, 2, unused, [pw_512], 6
+ lea dstq, [dstq+2*strideq]
+ VP9_IDCT8_WRITEx2 12, 13, 1, 2, unused, [pw_512], 6
+ lea dstq, [dstq+2*strideq]
+ VP9_IDCT8_WRITEx2 14, 15, 1, 2, unused, [pw_512], 6
+ lea dstq, [dstq+2*strideq]
+
+ ; at the end of the loop, m0 should still be zero
+ ; use that to zero out block coefficients
+ pxor m0, m0
+ ZERO_BLOCK blockq, 32, 16, m0
+ RET
+%endmacro
+%endif
+
+IADST16_YMM_FN idct, IDCT16, iadst, IADST16
+IADST16_YMM_FN iadst, IADST16, idct, IDCT16
+IADST16_YMM_FN iadst, IADST16, iadst, IADST16
+
;---------------------------------------------------------------------------------------------
; void vp9_idct_idct_32x32_add_<opt>(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob);
;---------------------------------------------------------------------------------------------
--
2.8.1
More information about the ffmpeg-devel
mailing list