highperfocused/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

x86/tx_float: add asm call versions of the 2pt and 4pt transforms

Browse Source 
Verified to be working.
This commit is contained in:
Lynne
2022-09-19 04:14:52 +02:00
parent 892548e6a1
commit 4ba68639ca
2 changed files with 32 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
libavutil/x86/tx_float.asm
View File

@@ -682,15 +682,27 @@ SECTION .text
%endmacro %endmacro
INIT_XMM sse3 INIT_XMM sse3
cglobal fft2_asm_float, 0, 0, 0, ctx, out, in, stride
movaps m0, [inq]
FFT2 m0, m1
movaps [outq], m0
add inq, mmsize*1
add outq, mmsize*1
ret
cglobal fft2_float, 4, 4, 2, ctx, out, in, stride cglobal fft2_float, 4, 4, 2, ctx, out, in, stride
movaps m0, [inq] movaps m0, [inq]
FFT2 m0, m1 FFT2 m0, m1
movaps [outq], m0 movaps [outq], m0
RET RET
%macro FFT4 2 %macro FFT4_FN 3
INIT_XMM sse2 INIT_XMM sse2
%if %3
cglobal fft4_ %+ %1 %+ _asm_float, 0, 0, 0, ctx, out, in, stride
%else
cglobal fft4_ %+ %1 %+ _float, 4, 4, 3, ctx, out, in, stride cglobal fft4_ %+ %1 %+ _float, 4, 4, 3, ctx, out, in, stride
%endif
movaps m0, [inq + 0*mmsize] movaps m0, [inq + 0*mmsize]
movaps m1, [inq + 1*mmsize] movaps m1, [inq + 1*mmsize]
@@ -708,11 +720,19 @@ cglobal fft4_ %+ %1 %+ _float, 4, 4, 3, ctx, out, in, stride
movaps [outq + 0*mmsize], m2 movaps [outq + 0*mmsize], m2
movaps [outq + 1*mmsize], m0 movaps [outq + 1*mmsize], m0
%if %3
add inq, mmsize*2
add outq, mmsize*2
ret
%else
RET RET
%endif
%endmacro %endmacro
FFT4 fwd, 0 FFT4_FN fwd, 0, 0
FFT4 inv, 1 FFT4_FN fwd, 0, 1
FFT4_FN inv, 1, 0
FFT4_FN inv, 1, 1
%macro FFT8_SSE_FN 1 %macro FFT8_SSE_FN 1
INIT_XMM sse3 INIT_XMM sse3

9
libavutil/x86/tx_float_init.c
View File

@@ -45,6 +45,9 @@ TX_DECL_FN(fft_sr_ns, avx2)
TX_DECL_FN(mdct_sr_inv, avx2) TX_DECL_FN(mdct_sr_inv, avx2)
TX_DECL_FN(fft2_asm, sse3)
TX_DECL_FN(fft4_fwd_asm, sse2)
TX_DECL_FN(fft4_inv_asm, sse2)
TX_DECL_FN(fft8_asm, sse3) TX_DECL_FN(fft8_asm, sse3)
TX_DECL_FN(fft8_asm, avx) TX_DECL_FN(fft8_asm, avx)
TX_DECL_FN(fft16_asm, avx) TX_DECL_FN(fft16_asm, avx)
@@ -101,8 +104,14 @@ static av_cold int m_inv_init(AVTXContext *s, const FFTXCodelet *cd,
const FFTXCodelet * const ff_tx_codelet_list_float_x86[] = { const FFTXCodelet * const ff_tx_codelet_list_float_x86[] = {
TX_DEF(fft2, FFT, 2, 2, 2, 0, 128, NULL, sse3, SSE3, AV_TX_INPLACE, 0), TX_DEF(fft2, FFT, 2, 2, 2, 0, 128, NULL, sse3, SSE3, AV_TX_INPLACE, 0),
TX_DEF(fft2_asm, FFT, 2, 2, 2, 0, 192, b8_i0, sse3, SSE3,
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, 0),
TX_DEF(fft2, FFT, 2, 2, 2, 0, 192, b8_i0, sse3, SSE3, AV_TX_INPLACE | FF_TX_PRESHUFFLE, 0), TX_DEF(fft2, FFT, 2, 2, 2, 0, 192, b8_i0, sse3, SSE3, AV_TX_INPLACE | FF_TX_PRESHUFFLE, 0),
TX_DEF(fft4_fwd, FFT, 4, 4, 2, 0, 128, NULL, sse2, SSE2, AV_TX_INPLACE | FF_TX_FORWARD_ONLY, 0), TX_DEF(fft4_fwd, FFT, 4, 4, 2, 0, 128, NULL, sse2, SSE2, AV_TX_INPLACE | FF_TX_FORWARD_ONLY, 0),
TX_DEF(fft4_fwd_asm, FFT, 4, 4, 2, 0, 192, b8_i0, sse2, SSE2,
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, 0),
TX_DEF(fft4_inv_asm, FFT, 4, 4, 2, 0, 128, NULL, sse2, SSE2,
AV_TX_INPLACE | FF_TX_INVERSE_ONLY | FF_TX_ASM_CALL, 0),
TX_DEF(fft4_fwd, FFT, 4, 4, 2, 0, 192, b8_i0, sse2, SSE2, AV_TX_INPLACE | FF_TX_PRESHUFFLE, 0), TX_DEF(fft4_fwd, FFT, 4, 4, 2, 0, 192, b8_i0, sse2, SSE2, AV_TX_INPLACE | FF_TX_PRESHUFFLE, 0),
TX_DEF(fft4_inv, FFT, 4, 4, 2, 0, 128, NULL, sse2, SSE2, AV_TX_INPLACE | FF_TX_INVERSE_ONLY, 0), TX_DEF(fft4_inv, FFT, 4, 4, 2, 0, 128, NULL, sse2, SSE2, AV_TX_INPLACE | FF_TX_INVERSE_ONLY, 0),
TX_DEF(fft8, FFT, 8, 8, 2, 0, 128, b8_i0, sse3, SSE3, AV_TX_INPLACE, 0), TX_DEF(fft8, FFT, 8, 8, 2, 0, 128, b8_i0, sse3, SSE3, AV_TX_INPLACE, 0),