lavc/h264dsp: R-V V 8-bit h264_idct_add16

While this *tends* to be faster than plain C, the performance numbers
are all over the place, presuambly due to the conditional character of
the main loop.

Some additional micro-optimisations should be feasible after the
underlying h264_idct_add and h264_idct_dc_add functions are also
implemented. Then it will no longer be necesseray to stricly abide by
the C ABI.

libavcodec/riscv/Makefile

@@ -31,7 +31,7 @@ RVV-OBJS-$(CONFIG_H263DSP) += riscv/h263dsp_rvv.o
 OBJS-$(CONFIG_H264CHROMA) += riscv/h264_chroma_init_riscv.o
 RVV-OBJS-$(CONFIG_H264CHROMA) += riscv/h264_mc_chroma.o
 OBJS-$(CONFIG_H264DSP) += riscv/h264dsp_init.o
-RVV-OBJS-$(CONFIG_H264DSP) += riscv/h264dsp_rvv.o
+RVV-OBJS-$(CONFIG_H264DSP) += riscv/h264dsp_rvv.o riscv/h264idct_rvv.o
 OBJS-$(CONFIG_HUFFYUV_DECODER) += riscv/huffyuvdsp_init.o
 RVV-OBJS-$(CONFIG_HUFFYUV_DECODER) += riscv/huffyuvdsp_rvv.o
 OBJS-$(CONFIG_IDCTDSP) += riscv/idctdsp_init.o

libavcodec/riscv/h264dsp_init.c

@@ -34,6 +34,10 @@ void ff_h264_h_loop_filter_luma_8_rvv(uint8_t *pix, ptrdiff_t stride,
 void ff_h264_h_loop_filter_luma_mbaff_8_rvv(uint8_t *pix, ptrdiff_t stride,
                                             int alpha, int beta, int8_t *tc0);
 
+void ff_h264_idct_add16_8_rvv(uint8_t *dst, const int *blockoffset,
+                              int16_t *block, int stride,
+                              const uint8_t nnzc[5 * 8]);
+
 extern int ff_startcode_find_candidate_rvb(const uint8_t *, int);
 extern int ff_startcode_find_candidate_rvv(const uint8_t *, int);
 
@@ -52,6 +56,10 @@ av_cold void ff_h264dsp_init_riscv(H264DSPContext *dsp, const int bit_depth,
             dsp->h264_h_loop_filter_luma = ff_h264_h_loop_filter_luma_8_rvv;
             dsp->h264_h_loop_filter_luma_mbaff =
                 ff_h264_h_loop_filter_luma_mbaff_8_rvv;
+
+#  if __riscv_xlen == 64
+            dsp->h264_idct_add16 = ff_h264_idct_add16_8_rvv;
+#  endif
         }
         dsp->startcode_find_candidate = ff_startcode_find_candidate_rvv;
     }

libavcodec/riscv/h264idct_rvv.S

@@ -0,0 +1,106 @@
+/*
+ * Copyright © 2024 Rémi Denis-Courmont.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "libavutil/riscv/asm.S"
+
+const ff_h264_scan8
+        .byte   014, 015, 024, 025, 016, 017, 026, 027
+        .byte   034, 035, 044, 045, 036, 037, 046, 047
+endconst
+
+#if (__riscv_xlen == 64)
+.irp    depth, 8
+func ff_h264_idct_add16_\depth\()_rvv, zve32x
+        addi    sp, sp, -80
+        lla     t0, ff_h264_scan8
+        sd      s0,   (sp)
+        li      t1, 32 << (\depth > 8)
+        mv      s0, sp
+        sd      ra,  8(sp)
+        sd      s1, 16(sp)
+        sd      s2, 24(sp)
+        sd      s3, 32(sp)
+        sd      s4, 40(sp)
+        sd      s5, 48(sp)
+        sd      s6, 56(sp)
+        sd      s7, 64(sp)
+        vsetivli  zero, 16, e8, m1, ta, ma
+        vle8.v    v8, (t0)
+        vlse16.v  v16, (a2), t1
+        vluxei8.v v12, (a4), v8
+.if \depth == 8
+        vsetvli   zero, zero, e16, m2, ta, ma
+.else
+        vsetvli   zero, zero, e32, m4, ta, ma
+.endif
+        vmsne.vi  v1, v16, 0
+        vsetvli   zero, zero, e8, m1, ta, ma
+        vmseq.vi  v2, v12, 1
+        vmsne.vi  v0, v12, 0
+        vmand.mm  v1, v1, v2
+        vsetvli   zero, zero, e16, m2, ta, ma
+        vmv.x.s   s2, v0
+        vmv.x.s   s3, v1
+        li      s1, 16
+        mv      s4, a0
+        mv      s5, a1
+        mv      s6, a2
+        mv      s7, a3
+1:
+        andi    t0, s2, 1
+        addi    s1, s1, -1
+        srli    s2, s2, 1
+        beqz    t0, 3f     # if (nnz)
+        lw      t2, (s5)   # block_offset[i]
+        andi    t1, s3, 1
+        mv      a1, s6
+        mv      a2, s7
+        add     a0, s4, t2
+        beqz    t1, 2f    # if (nnz == 1 && block[i * 16])
+        call    ff_h264_idct_dc_add_\depth\()_c
+        j       3f
+2:
+        call    ff_h264_idct_add_\depth\()_c
+3:
+        srli    s3, s3, 1
+        addi    s5, s5, 4
+        addi    s6, s6, 16 * 2 << (\depth > 8)
+        bnez    s1, 1b
+
+        ld      s7, 64(sp)
+        ld      s6, 56(sp)
+        ld      s5, 48(sp)
+        ld      s4, 40(sp)
+        ld      s3, 32(sp)
+        ld      s2, 24(sp)
+        ld      s1, 16(sp)
+        ld      ra,  8(sp)
+        ld      s0,  0(sp)
+        addi    sp, sp, 80
+        ret
+endfunc
+.endr
+#endif