Cycle 3 (MC interpolation) closure: M1'''=100%, R'''=0.067 RED, M4=-19.5%
Third daedalus-fourier kernel — VP9 8-tap regular subpel filter,
horizontal direction, 8-wide output. Multiply-heavy by design to
stress V3D's no-DP4A deficit. Full cycle Phase 1-7 + M4'''.
Phase 5''' second-model review delivered cleanly — caught 1 RED
bug pre-implementation (src_off off-by-3 indexing convention) and
2 YELLOW gaps (assert MUST language, shaderdb filter-LUT gate).
Without the review, M1''' would have failed silently on first run
with cryptic "high-index source pixels wrong" symptoms.
Phase 6 v1 first-light: M1''' 100.0000% bit-exact (65536/65536
blocks across all 16 mx phases). Phase 5''' filter-LUT prediction
materialised exactly: 197 uniforms (gate was 144), 2 threads (down
from cycle-2's 4 due to register pressure).
Performance:
M2''' = 1.413 Mblock/s (707.9 ns/block)
M3''' = 20.997 Mblock/s (NEON baseline phase3)
R''' = 0.067 (RED band — structural mismatch)
shaderdb: 488 inst, 2 threads, 197 uniforms, 25 max-temps, 0 spills
M4''' concurrent matrix (8s windows):
NEON 1-core 14.479 Mblock/s
NEON 4-core 15.248 Mblock/s <- baseline (compute-bound,
not bandwidth-saturated
like cycles 1+2!)
QPU only 1.380 Mblock/s
MIXED NEON-3 + QPU 12.277 Mblock/s <- -19.5% (FAIL gate)
MIXED NEON-4 + QPU 12.158 Mblock/s <- -20.3%
NEW cross-cycle finding (Phase 9 lesson 2): compute-bound CPU
workloads make the QPU-offload story collapse. Cycles 1+2 were
bandwidth-saturated (4-core scaling 0.56-0.82x of 1-core), so
freeing a CPU core via QPU offload added throughput. Cycle 3 MC
is compute-bound (4-core scaling 1.05x of 1-core — near-linear),
no free cycles to free. QPU contribution (0.45 Mblock/s in
contention) doesn't compensate for losing 1 NEON core delivering
~3.8 Mblock/s.
But 30fps@1080p floor: PASS in every config (1.4x to 15.7x
isolation margin). Per project_30fps_floor_is_fine.md, user-facing
test never fails — daily YouTube playback works fine on any CPU/QPU
split.
DEPLOYMENT RECIPE for higgs (cycle 3 confirmed split):
IDCT (k1) -> QPU (R=0.92, +7% mixed, frees CPU core)
LPF (k2) -> QPU (R=0.41, +7% mixed, frees CPU core)
MC (k3) -> CPU (R=0.067, -19.5% mixed — stays on CPU)
Entropy -> CPU (structurally serial)
Mixed-substrate deployment, not "QPU does everything". Realistic for
higgs: entropy + MC on 2-3 ARM cores; IDCT + LPF dispatched to QPU
concurrently; 1-2 ARM cores left for vscode etc.
New artifacts:
- src/v3d_mc_8h.comp — GLSL kernel
- tests/vp9_mc_ref.c — standalone C ref (REGULAR filter
embedded; clean transcription)
- tests/bench_neon_mc.c — M1'''_c + M3''' bench
- tests/bench_v3d_mc.c — M1''' + M2''' bench with contract
asserts + 30fps margin display
- tests/bench_concurrent_mc.c — M4''' pthread bench
- external/ffmpeg-snapshot/libavcodec/aarch64/vp9mc_neon.S (vendored)
- external/ffmpeg-snapshot/libavcodec/vp9_subpel_filters_table.c
(hand-extracted; provides
ff_vp9_subpel_filters symbol
without dragging in full vp9dsp.c)
- docs/k3_mc_phase{1,2,3,4,5,7}.md — full cycle documentation
Memory updates: project_30fps_floor_is_fine.md (user's 30fps target
recalibration), MEMORY.md index updated.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -0,0 +1,220 @@
|
||||
/*
|
||||
* Cycle 3 Phase 3 — NEON M3''' baseline for VP9 8-tap regular
|
||||
* horizontal MC interpolation, 8×8 block.
|
||||
*
|
||||
* Reports:
|
||||
* M1'''_c (correctness): C-ref ↔ NEON bit-exact rate, N random
|
||||
* 8×8 blocks with random source pixels and
|
||||
* random subpel phase mx ∈ [0, 15]
|
||||
* M3''' (throughput): NEON sustained Mblock/s, single-thread,
|
||||
* time-based
|
||||
*
|
||||
* License: LGPL-2.1+ (statically links FFmpeg NEON snapshot).
|
||||
*/
|
||||
#define _POSIX_C_SOURCE 200809L
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include <stdint.h>
|
||||
#include <stddef.h>
|
||||
#include <string.h>
|
||||
#include <time.h>
|
||||
#include <getopt.h>
|
||||
|
||||
extern void daedalus_vp9_put_regular_8h_ref(
|
||||
uint8_t *dst, ptrdiff_t dst_stride,
|
||||
const uint8_t *src, ptrdiff_t src_stride,
|
||||
int h, int mx, int my);
|
||||
|
||||
extern void ff_vp9_put_regular8_h_neon(
|
||||
uint8_t *dst, ptrdiff_t dst_stride,
|
||||
const uint8_t *src, ptrdiff_t src_stride,
|
||||
int h, int mx, int my);
|
||||
|
||||
/* RNG ------------------------------------------------------------ */
|
||||
|
||||
static uint64_t xs_state;
|
||||
static inline uint64_t xs(void) {
|
||||
uint64_t x = xs_state;
|
||||
x ^= x << 13; x ^= x >> 7; x ^= x << 17;
|
||||
return xs_state = x;
|
||||
}
|
||||
|
||||
/* Block layout: each block gets its own 8×16 source buffer + 8×8 dst.
|
||||
* - source buffer is 16 cols wide; the filter is called with
|
||||
* src = block_src + 3, so it reads cols [src+0-3..src+8+4] =
|
||||
* [0..14] of the 16-col buffer. col 15 is unused padding.
|
||||
* - dst is 8 cols × 8 rows.
|
||||
*/
|
||||
#define SRC_W 16
|
||||
#define SRC_H 8
|
||||
#define DST_W 8
|
||||
#define DST_H 8
|
||||
#define SRC_BYTES (SRC_H * SRC_W) /* 128 */
|
||||
#define DST_BYTES (DST_H * DST_W) /* 64 */
|
||||
|
||||
static void gen_src(uint8_t *buf)
|
||||
{
|
||||
for (int i = 0; i < SRC_BYTES; i++)
|
||||
buf[i] = (uint8_t)(xs() & 0xff);
|
||||
}
|
||||
|
||||
static double now_seconds(void)
|
||||
{
|
||||
struct timespec ts;
|
||||
clock_gettime(CLOCK_MONOTONIC_RAW, &ts);
|
||||
return ts.tv_sec + ts.tv_nsec * 1e-9;
|
||||
}
|
||||
|
||||
/* M1'''_c correctness gate -------------------------------------- */
|
||||
|
||||
static int correctness_check(uint64_t seed, int n_blocks)
|
||||
{
|
||||
xs_state = seed ? seed : 0xabcdef1234567890ULL;
|
||||
int mismatches = 0;
|
||||
uint8_t src[SRC_BYTES];
|
||||
uint8_t dst_a[DST_BYTES], dst_b[DST_BYTES];
|
||||
|
||||
int mx_hist[16] = {0};
|
||||
|
||||
for (int i = 0; i < n_blocks; i++) {
|
||||
gen_src(src);
|
||||
int mx = (int)(xs() & 15);
|
||||
mx_hist[mx]++;
|
||||
|
||||
memset(dst_a, 0, DST_BYTES);
|
||||
memset(dst_b, 0, DST_BYTES);
|
||||
|
||||
daedalus_vp9_put_regular_8h_ref(dst_a, DST_W, src + 3, SRC_W, DST_H, mx, 0);
|
||||
ff_vp9_put_regular8_h_neon (dst_b, DST_W, src + 3, SRC_W, DST_H, mx, 0);
|
||||
|
||||
if (memcmp(dst_a, dst_b, DST_BYTES) != 0) {
|
||||
if (mismatches < 3) {
|
||||
fprintf(stderr, "MISMATCH block %d mx=%d:\n", i, mx);
|
||||
fprintf(stderr, " ref:");
|
||||
for (int r = 0; r < 8; r++) {
|
||||
fprintf(stderr, "\n r%d ", r);
|
||||
for (int c = 0; c < 8; c++) fprintf(stderr, "%3u ", dst_a[r*8+c]);
|
||||
}
|
||||
fprintf(stderr, "\n neon:");
|
||||
for (int r = 0; r < 8; r++) {
|
||||
fprintf(stderr, "\n r%d ", r);
|
||||
for (int c = 0; c < 8; c++) fprintf(stderr, "%3u ", dst_b[r*8+c]);
|
||||
}
|
||||
fprintf(stderr, "\n");
|
||||
}
|
||||
mismatches++;
|
||||
}
|
||||
}
|
||||
printf("M1'''_c correctness: %d / %d blocks bit-exact (%.4f%%)\n",
|
||||
n_blocks - mismatches, n_blocks,
|
||||
100.0 * (n_blocks - mismatches) / n_blocks);
|
||||
/* mx histogram — confirms all 16 phases get exercised. */
|
||||
int min_mx = mx_hist[0], max_mx = mx_hist[0];
|
||||
for (int i = 1; i < 16; i++) {
|
||||
if (mx_hist[i] < min_mx) min_mx = mx_hist[i];
|
||||
if (mx_hist[i] > max_mx) max_mx = mx_hist[i];
|
||||
}
|
||||
printf(" mx phase coverage: min=%d max=%d (16 phases sampled)\n",
|
||||
min_mx, max_mx);
|
||||
return mismatches;
|
||||
}
|
||||
|
||||
/* M3''' throughput ---------------------------------------------- */
|
||||
|
||||
static void throughput_neon(uint64_t seed, int n_blocks, double duration_s)
|
||||
{
|
||||
xs_state = seed ? seed : 0xdeadbeef12345678ULL;
|
||||
|
||||
uint8_t *master_src = malloc((size_t) n_blocks * SRC_BYTES);
|
||||
uint8_t *work_src = malloc((size_t) n_blocks * SRC_BYTES);
|
||||
uint8_t *dsts = malloc((size_t) n_blocks * DST_BYTES);
|
||||
int *mxs = malloc(n_blocks * sizeof(int));
|
||||
if (!master_src || !work_src || !dsts || !mxs) { fprintf(stderr, "alloc fail\n"); exit(1); }
|
||||
|
||||
for (int i = 0; i < n_blocks; i++) {
|
||||
gen_src(master_src + (size_t)i * SRC_BYTES);
|
||||
mxs[i] = (int)(xs() & 15);
|
||||
}
|
||||
|
||||
/* Warm. */
|
||||
memcpy(work_src, master_src, (size_t) n_blocks * SRC_BYTES);
|
||||
for (int i = 0; i < n_blocks; i++)
|
||||
ff_vp9_put_regular8_h_neon(dsts + (size_t)i * DST_BYTES, DST_W,
|
||||
work_src + (size_t)i * SRC_BYTES + 3, SRC_W,
|
||||
DST_H, mxs[i], 0);
|
||||
|
||||
double t0 = now_seconds();
|
||||
double t_end = t0 + duration_s;
|
||||
uint64_t done = 0;
|
||||
while (now_seconds() < t_end) {
|
||||
memcpy(work_src, master_src, (size_t) n_blocks * SRC_BYTES);
|
||||
for (int i = 0; i < n_blocks; i++)
|
||||
ff_vp9_put_regular8_h_neon(dsts + (size_t)i * DST_BYTES, DST_W,
|
||||
work_src + (size_t)i * SRC_BYTES + 3, SRC_W,
|
||||
DST_H, mxs[i], 0);
|
||||
done += n_blocks;
|
||||
}
|
||||
double elapsed = now_seconds() - t0;
|
||||
|
||||
/* setup-only subtraction */
|
||||
int setup_iters = (int) (done / n_blocks);
|
||||
double s0 = now_seconds();
|
||||
for (int it = 0; it < setup_iters; it++)
|
||||
memcpy(work_src, master_src, (size_t) n_blocks * SRC_BYTES);
|
||||
double s1 = now_seconds();
|
||||
|
||||
double kernel_seconds = elapsed - (s1 - s0);
|
||||
double mbps = done / kernel_seconds / 1e6;
|
||||
|
||||
printf("M3''' NEON throughput:\n");
|
||||
printf(" blocks/batch: %d\n", n_blocks);
|
||||
printf(" batches done: %d\n", setup_iters);
|
||||
printf(" total blocks: %llu\n", (unsigned long long) done);
|
||||
printf(" elapsed (kernel)=%.6f s\n", kernel_seconds);
|
||||
printf(" elapsed (setup) =%.6f s\n", s1 - s0);
|
||||
printf(" throughput = %.3f Mblock/s\n", mbps);
|
||||
printf(" per-block = %.1f ns\n", kernel_seconds / done * 1e9);
|
||||
/* 1080p: 32400 blocks/frame */
|
||||
printf(" equiv 1080p = %.1f FPS (32400 blocks/frame)\n",
|
||||
mbps * 1e6 / 32400.0);
|
||||
|
||||
free(master_src); free(work_src); free(dsts); free(mxs);
|
||||
}
|
||||
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
int n_blocks = 65536;
|
||||
double duration = 5.0;
|
||||
uint64_t seed = 0;
|
||||
int do_correctness = 1;
|
||||
|
||||
static struct option opts[] = {
|
||||
{"blocks", required_argument, 0, 'b'},
|
||||
{"duration", required_argument, 0, 'd'},
|
||||
{"seed", required_argument, 0, 's'},
|
||||
{"no-correctness", no_argument, 0, 'C'},
|
||||
{0,0,0,0}
|
||||
};
|
||||
for (int c; (c = getopt_long(argc, argv, "b:d:s:C", opts, 0)) != -1;) {
|
||||
switch (c) {
|
||||
case 'b': n_blocks = atoi(optarg); break;
|
||||
case 'd': duration = atof(optarg); break;
|
||||
case 's': seed = strtoull(optarg, 0, 0); break;
|
||||
case 'C': do_correctness = 0; break;
|
||||
default: return 2;
|
||||
}
|
||||
}
|
||||
|
||||
if (do_correctness) {
|
||||
printf("=== M1'''_c bit-exact (10000 random blocks) ===\n");
|
||||
if (correctness_check(seed, 10000) != 0) {
|
||||
fprintf(stderr, "REFUSING to measure throughput on a broken kernel.\n");
|
||||
return 1;
|
||||
}
|
||||
printf("\n");
|
||||
}
|
||||
|
||||
printf("=== M3''' NEON throughput ===\n");
|
||||
throughput_neon(seed, n_blocks, duration);
|
||||
return 0;
|
||||
}
|
||||
Reference in New Issue
Block a user