[FFmpeg-devel] [PATCH 2/2] lavc/flacdsp: R-V V LPC32

[Rémi Denis-Courmont]

The entire set of 32 coefficients and corresponding past 32 samples can
fit in a single vector (with LMUL=8) exactly, but... since widening
double the needed vector sizes, we still end up too short with 128-bit
vectors. This adds a very simple version for future 256+-bit hardware,
and for pred_orders values up to 16, and a bit more involved loop for
for 128-bit hardware with pred_orders between 17 and 32.

With 128-bit hardware, the benchmarks look like this:
flac_lpc_32_13_c:       30152.0
flac_lpc_32_13_rvv_i32: 10244.7
flac_lpc_32_16_c:       37314.2
flac_lpc_32_16_rvv_i32: 10126.2
flac_lpc_32_29_c:       61910.0
flac_lpc_32_29_rvv_i32: 14495.2
flac_lpc_32_32_c:       68204.0
flac_lpc_32_32_rvv_i32: 13273.7
---
 libavcodec/riscv/flacdsp_init.c | 12 +++++++
 libavcodec/riscv/flacdsp_rvv.S  | 57 +++++++++++++++++++++++++++++++++
 2 files changed, 69 insertions(+)

diff --git a/libavcodec/riscv/flacdsp_init.c b/libavcodec/riscv/flacdsp_init.c
index 73d431cb77..f60f98ea31 100644
--- a/libavcodec/riscv/flacdsp_init.c
+++ b/libavcodec/riscv/flacdsp_init.c
@@ -22,8 +22,13 @@
 
 #include "libavutil/attributes.h"
 #include "libavutil/cpu.h"
+#include "libavutil/riscv/cpu.h"
 #include "libavcodec/flacdsp.h"
 
+void ff_flac_lpc32_rvv(int32_t *decoded, const int coeffs[32],
+                       int pred_order, int qlevel, int len);
+void ff_flac_lpc32_rvv_simple(int32_t *decoded, const int coeffs[32],
+                              int pred_order, int qlevel, int len);
 void ff_flac_decorrelate_indep2_16_rvv(uint8_t **out, int32_t **in,
                                        int channels, int len, int shift);
 void ff_flac_decorrelate_indep4_16_rvv(uint8_t **out, int32_t **in,
@@ -60,6 +65,13 @@ av_cold void ff_flacdsp_init_riscv(FLACDSPContext *c, enum AVSampleFormat fmt,
     int flags = av_get_cpu_flags();
 
     if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB_ADDR)) {
+        int vlenb = ff_get_rv_vlenb();
+
+        if (vlenb == 16)
+            c->lpc32 = ff_flac_lpc32_rvv;
+        else if (vlenb > 16) 
+            c->lpc32 = ff_flac_lpc32_rvv_simple;
+
         switch (fmt) {
         case AV_SAMPLE_FMT_S16:
             switch (channels) {
diff --git a/libavcodec/riscv/flacdsp_rvv.S b/libavcodec/riscv/flacdsp_rvv.S
index 12b456f7da..b1724f5500 100644
--- a/libavcodec/riscv/flacdsp_rvv.S
+++ b/libavcodec/riscv/flacdsp_rvv.S
@@ -21,6 +21,63 @@
 #include "libavutil/riscv/asm.S"
 
 #if (__riscv_xlen == 64)
+func ff_flac_lpc32_rvv, zve32x
+        addi    t2, a2, -16
+        ble     t2, zero, ff_flac_lpc32_rvv_simple
+        vsetivli zero, 1, e64, m1, ta, ma
+        vmv.s.x v0, zero
+        vsetvli zero, a2, e32, m8, ta, ma
+        vle32.v v8, (a1)
+        sub     a4, a4, a2
+        vle32.v v16, (a0)
+        sh2add  a0, a2, a0
+1:
+        vsetvli zero, a2, e32, m4, ta, ma
+        vwmul.vv v24, v8, v16
+        vsetvli zero, t2, e32, m4, tu, ma
+        vwmacc.vv v24, v12, v20
+        vsetvli zero, a2, e64, m8, ta, ma
+        vredsum.vs v24, v24, v0
+        lw      t0, (a0)
+        addi    a4, a4, -1
+        vmv.x.s t1, v24
+        vsetvli zero, a2, e32, m8, ta, ma
+        sra     t1, t1, a3
+        add     t0, t0, t1
+        vslide1down.vx v16, v16, t0
+        sw      t0, (a0)
+        addi    a0, a0, 4
+        bnez    a4, 1b
+
+        ret
+endfunc
+
+func ff_flac_lpc32_rvv_simple, zve32x
+        vsetivli zero, 1, e64, m1, ta, ma
+        vmv.s.x v0, zero
+        vsetvli zero, a2, e32, m4, ta, ma
+        vle32.v v8, (a1)
+        sub     a4, a4, a2
+        vle32.v v16, (a0)
+        sh2add  a0, a2, a0
+1:
+        vwmul.vv v24, v8, v16
+        vsetvli zero, zero, e64, m8, ta, ma
+        vredsum.vs v24, v24, v0
+        lw      t0, (a0)
+        addi    a4, a4, -1
+        vmv.x.s t1, v24
+        vsetvli zero, zero, e32, m4, ta, ma
+        sra     t1, t1, a3
+        add     t0, t0, t1
+        vslide1down.vx v16, v16, t0
+        sw      t0, (a0)
+        addi    a0, a0, 4
+        bnez    a4, 1b
+
+        ret
+endfunc
+
 func ff_flac_decorrelate_indep2_16_rvv, zve32x
         ld      a0,  (a0)
         ld      a2, 8(a1)
-- 
2.42.0