FFmpeg/libavcodec/riscv/Makefile

OBJS-$(CONFIG_AAC_DECODER) += riscv/aacpsdsp_init.o
RVV-OBJS-$(CONFIG_AAC_DECODER) += riscv/aacpsdsp_rvv.o
OBJS-$(CONFIG_ALAC_DECODER) += riscv/alacdsp_init.o
RVV-OBJS-$(CONFIG_ALAC_DECODER) += riscv/alacdsp_rvv.o
OBJS-$(CONFIG_AUDIODSP) += riscv/audiodsp_init.o \
                           riscv/audiodsp_rvf.o
RVV-OBJS-$(CONFIG_AUDIODSP) += riscv/audiodsp_rvv.o
OBJS-$(CONFIG_BSWAPDSP) += riscv/bswapdsp_init.o \
                           riscv/bswapdsp_rvb.o
RVV-OBJS-$(CONFIG_BSWAPDSP) += riscv/bswapdsp_rvv.o
OBJS-$(CONFIG_FMTCONVERT) += riscv/fmtconvert_init.o
RVV-OBJS-$(CONFIG_FMTCONVERT) += riscv/fmtconvert_rvv.o
OBJS-$(CONFIG_IDCTDSP) += riscv/idctdsp_init.o
RVV-OBJS-$(CONFIG_IDCTDSP) += riscv/idctdsp_rvv.o
OBJS-$(CONFIG_OPUS_DECODER) += riscv/opusdsp_init.o
RVV-OBJS-$(CONFIG_OPUS_DECODER) += riscv/opusdsp_rvv.o
OBJS-$(CONFIG_PIXBLOCKDSP) += riscv/pixblockdsp_init.o \
                              riscv/pixblockdsp_rvi.o
RVV-OBJS-$(CONFIG_PIXBLOCKDSP) += riscv/pixblockdsp_rvv.o
OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_init.o
RVV-OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_rvv.o
lavc/aacpsdsp: RISC-V V add_squares 2022-09-26 17:52:46 +03:00			`OBJS-$(CONFIG_AAC_DECODER) += riscv/aacpsdsp_init.o`
			`RVV-OBJS-$(CONFIG_AAC_DECODER) += riscv/aacpsdsp_rvv.o`
lavc/alacdsp: RISC-V V decorrelate_stereo To avoid data dependencies, this does the following unroll, which requires one extra but probably free addition: coeff = (b * left_weight) >> decorr_shift; b += a; a -= coeff; b -= coeff; swap(a, b); 2022-10-04 20:17:16 +03:00			`OBJS-$(CONFIG_ALAC_DECODER) += riscv/alacdsp_init.o`
			`RVV-OBJS-$(CONFIG_ALAC_DECODER) += riscv/alacdsp_rvv.o`
lavc/audiodsp: RISC-V F vector_clipf RV64G supports MIN & MAX instructions natively only on floating point registers, not general purpose ones. The later would require the Zbb extension. Due to that, it is actually faster to perform the clipping "properly" in FPU. Benchmarks on SiFive U74-MC (courtesy of Shanghai StarFive Tech): audiodsp.vector_clipf_c: 29551.5 audiodsp.vector_clipf_rvf: 17871.0 Also tried unrolling with 2 or 8 elements but it gets worse either way. 2022-09-26 17:52:23 +03:00			`OBJS-$(CONFIG_AUDIODSP) += riscv/audiodsp_init.o \`
			`riscv/audiodsp_rvf.o`
lavc/audiodsp: RISC-V V vector_clip_int32 2022-09-26 17:52:40 +03:00			`RVV-OBJS-$(CONFIG_AUDIODSP) += riscv/audiodsp_rvv.o`
lavc/bswapdsp: RISC-V B bswap_buf Simply taking the Zbb REV8 instruction into use in a simple loop gives some significant savings: bswap_buf_c: 1081.0 bswap_buf_rvb_b: 771.0 But we can also use the 64-bit REV8 as a pseudo-SIMD instruction with just one additional shift, and one fewer load, effectively doubling the bandwidth. Consequently, this patch is useful even if the compile-time target has Zbb enabled for C code: bswap_buf_c: 1081.0 bswap_buf_rvb_b: 341.0 (this patch) On the other hand, this approach fails miserably for bswap16_buf as the ratio of shifts and stores becomes unfavorable compared to naïve C: bswap16_buf_c: 1542.0 bswap16_buf_rvb_b: 1803.7 Unrolling to process 128 bits (4 samples) at a time actually worsens performance ever so slightly: bswap_buf_c: 1081.0 bswap_buf_rvb_b: 408.5 2022-10-02 14:54:59 +03:00			`OBJS-$(CONFIG_BSWAPDSP) += riscv/bswapdsp_init.o \`
			`riscv/bswapdsp_rvb.o`
lavc/bswapdsp: RISC-V V bswap_buf 2022-10-02 14:55:00 +03:00			`RVV-OBJS-$(CONFIG_BSWAPDSP) += riscv/bswapdsp_rvv.o`
lavc/fmtconvert: RISC-V V int32_to_float_fmul_scalar 2022-09-26 17:52:43 +03:00			`OBJS-$(CONFIG_FMTCONVERT) += riscv/fmtconvert_init.o`
			`RVV-OBJS-$(CONFIG_FMTCONVERT) += riscv/fmtconvert_rvv.o`
lavc/idctdsp: RISC-V V put_pixels_clamped function 2022-09-27 23:04:22 +03:00			`OBJS-$(CONFIG_IDCTDSP) += riscv/idctdsp_init.o`
			`RVV-OBJS-$(CONFIG_IDCTDSP) += riscv/idctdsp_rvv.o`
lavc/opusdsp: RISC-V V (128-bit) postfilter This is implemented for a vector size of 128-bit. Since the scalar product in the inner loop covers 5 samples or 160 bits, we need a group multipler of 2. To avoid reconfiguring the vector type, the outer loop, which loads multiple input samples sticks to the same multipler. Consequently, the outer loop loads 8 samples per iteration. This is safe since the minimum period of the CELT codec is 15 samples. The same code would also work, albeit needlessly inefficiently with a vector length of 256 bits. A proper implementation will follow instead. 2022-10-05 19:12:53 +03:00			`OBJS-$(CONFIG_OPUS_DECODER) += riscv/opusdsp_init.o`
			`RVV-OBJS-$(CONFIG_OPUS_DECODER) += riscv/opusdsp_rvv.o`
lavc/pixblockdsp: RISC-V I get_pixels Benchmarks on SiFive U74-MC (courtesy of Shanghai StarFive Tech): get_pixels_c: 180.0 get_pixels_rvi: 136.7 2022-09-26 17:52:24 +03:00			`OBJS-$(CONFIG_PIXBLOCKDSP) += riscv/pixblockdsp_init.o \`
			`riscv/pixblockdsp_rvi.o`
lavc/pixblockdsp: RISC-V V 8-bit get_pixels & get_pixels_unaligned 2022-09-27 23:04:25 +03:00			`RVV-OBJS-$(CONFIG_PIXBLOCKDSP) += riscv/pixblockdsp_rvv.o`
lavc/vorbisdsp: RISC-V V inverse_coupling This uses the following vectorisation: for (i = 0; i < blocksize; i++) { ang[i] = mag[i] - copysignf(fmaxf(ang[i], 0.f), mag[i]); mag[i] = mag[i] - copysignf(fminf(ang[i], 0.f), mag[i]); } 2022-09-26 17:52:45 +03:00			`OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_init.o`
			`RVV-OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_rvv.o`