diff options
author | Ronald S. Bultje <rsbultje@gmail.com> | 2016-07-09 00:51:56 +0300 |
---|---|---|
committer | Ronald S. Bultje <rsbultje@gmail.com> | 2016-07-11 17:14:58 +0300 |
commit | f0a2b6249bb2426befa4c03247361268e45b13af (patch) | |
tree | f86bf79a94cef01c09081900d6effe66811eaa1f /libavutil/x86/x86util.asm | |
parent | a2550e7d2e4c1e2f1fd262985a1d789240f7e839 (diff) |
vp9: add 16x16 idct avx2 (8-bit).
checkasm --bench, 10k runs, for *_add_${bpc}_${sub_idct}_${opt}, shows
that it's about 1.65x as fast as the AVX version for the full IDCT, and
similar speedups for the sub-IDCTs:
nop: 24.6
vp9_inv_dct_dct_16x16_add_8_1_c: 6444.8
vp9_inv_dct_dct_16x16_add_8_1_sse2: 638.6
vp9_inv_dct_dct_16x16_add_8_1_ssse3: 484.4
vp9_inv_dct_dct_16x16_add_8_1_avx: 661.2
vp9_inv_dct_dct_16x16_add_8_1_avx2: 311.5
vp9_inv_dct_dct_16x16_add_8_2_c: 6665.7
vp9_inv_dct_dct_16x16_add_8_2_sse2: 646.9
vp9_inv_dct_dct_16x16_add_8_2_ssse3: 455.2
vp9_inv_dct_dct_16x16_add_8_2_avx: 521.9
vp9_inv_dct_dct_16x16_add_8_2_avx2: 304.3
vp9_inv_dct_dct_16x16_add_8_4_c: 7022.7
vp9_inv_dct_dct_16x16_add_8_4_sse2: 647.4
vp9_inv_dct_dct_16x16_add_8_4_ssse3: 467.1
vp9_inv_dct_dct_16x16_add_8_4_avx: 446.1
vp9_inv_dct_dct_16x16_add_8_4_avx2: 297.0
vp9_inv_dct_dct_16x16_add_8_8_c: 6800.4
vp9_inv_dct_dct_16x16_add_8_8_sse2: 598.6
vp9_inv_dct_dct_16x16_add_8_8_ssse3: 465.7
vp9_inv_dct_dct_16x16_add_8_8_avx: 440.9
vp9_inv_dct_dct_16x16_add_8_8_avx2: 290.2
vp9_inv_dct_dct_16x16_add_8_16_c: 6626.6
vp9_inv_dct_dct_16x16_add_8_16_sse2: 599.5
vp9_inv_dct_dct_16x16_add_8_16_ssse3: 475.0
vp9_inv_dct_dct_16x16_add_8_16_avx: 469.9
vp9_inv_dct_dct_16x16_add_8_16_avx2: 286.4
Diffstat (limited to 'libavutil/x86/x86util.asm')
-rw-r--r-- | libavutil/x86/x86util.asm | 69 |
1 files changed, 68 insertions, 1 deletions
diff --git a/libavutil/x86/x86util.asm b/libavutil/x86/x86util.asm index e7493f4883..44ed750ae5 100644 --- a/libavutil/x86/x86util.asm +++ b/libavutil/x86/x86util.asm @@ -30,7 +30,10 @@ %include "libavutil/x86/x86inc.asm" %macro SBUTTERFLY 4 -%if avx_enabled == 0 +%ifidn %1, dqqq + vperm2i128 m%4, m%2, m%3, q0301 + vinserti128 m%2, m%2, xm%3, 1 +%elif avx_enabled == 0 mova m%4, m%2 punpckl%1 m%2, m%3 punpckh%1 m%4, m%3 @@ -193,6 +196,70 @@ %endif %endmacro +%macro TRANSPOSE16x16W 18-19 +; in: m0..m15, unless %19 in which case m6 is in %17 +; out: m0..m15, unless %19 in which case m4 is in %18 +; spills into %17 and %18 +%if %0 < 19 + mova %17, m%7 +%endif + + SBUTTERFLY dqqq, %1, %9, %7 + SBUTTERFLY dqqq, %2, %10, %7 + SBUTTERFLY dqqq, %3, %11, %7 + SBUTTERFLY dqqq, %4, %12, %7 + SBUTTERFLY dqqq, %5, %13, %7 + SBUTTERFLY dqqq, %6, %14, %7 + mova %18, m%14 + mova m%7, %17 + SBUTTERFLY dqqq, %7, %15, %14 + SBUTTERFLY dqqq, %8, %16, %14 + + SBUTTERFLY wd, %1, %2, %14 + SBUTTERFLY wd, %3, %4, %14 + SBUTTERFLY wd, %5, %6, %14 + SBUTTERFLY wd, %7, %8, %14 + SBUTTERFLY wd, %9, %10, %14 + SBUTTERFLY wd, %11, %12, %14 + mova %17, m%12 + mova m%14, %18 + SBUTTERFLY wd, %13, %14, %12 + SBUTTERFLY wd, %15, %16, %12 + + SBUTTERFLY dq, %1, %3, %12 + SBUTTERFLY dq, %2, %4, %12 + SBUTTERFLY dq, %5, %7, %12 + SBUTTERFLY dq, %6, %8, %12 + SBUTTERFLY dq, %9, %11, %12 + mova %18, m%11 + mova m%12, %17 + SBUTTERFLY dq, %10, %12, %11 + SBUTTERFLY dq, %13, %15, %11 + SBUTTERFLY dq, %14, %16, %11 + + SBUTTERFLY qdq, %1, %5, %11 + SBUTTERFLY qdq, %2, %6, %11 + SBUTTERFLY qdq, %3, %7, %11 + SBUTTERFLY qdq, %4, %8, %11 + + SWAP %2, %5 + SWAP %4, %7 + + SBUTTERFLY qdq, %9, %13, %11 + SBUTTERFLY qdq, %10, %14, %11 + mova m%11, %18 + mova %18, m%5 + SBUTTERFLY qdq, %11, %15, %5 + SBUTTERFLY qdq, %12, %16, %5 + +%if %0 < 19 + mova m%5, %18 +%endif + + SWAP %10, %13 + SWAP %12, %15 +%endmacro + ; PABSW macro assumes %1 != %2, while ABS1/2 macros work in-place %macro PABSW 2 %if cpuflag(ssse3) |