Commit 47a581a5 authored by Ronald S. Bultje's avatar Ronald S. Bultje

Change emu_edge function argument order for easier SIMD

parent 404a8fbf
......@@ -87,9 +87,8 @@ void (name)(pixel *dst, ptrdiff_t dst_stride, \
typedef decl_blend_fn(*blend_fn);
#define decl_emu_edge_fn(name) \
void (name)(pixel *dst, ptrdiff_t dst_stride, \
const pixel *src, ptrdiff_t src_stride, \
int bw, int bh, int iw, int ih, int x, int y)
void (name)(intptr_t bw, intptr_t bh, intptr_t iw, intptr_t ih, intptr_t x, intptr_t y, \
pixel *dst, ptrdiff_t dst_stride, const pixel *src, ptrdiff_t src_stride)
typedef decl_emu_edge_fn(*emu_edge_fn);
typedef struct Dav1dMCDSPContext {
......
......@@ -521,11 +521,11 @@ static void warp_affine_8x8t_c(coef *tmp, const ptrdiff_t tmp_stride,
}
}
static void emu_edge_c(pixel *dst, const ptrdiff_t dst_stride,
const pixel *ref, const ptrdiff_t ref_stride,
const int bw, const int bh,
const int iw, const int ih,
const int x, const int y)
static void emu_edge_c(const intptr_t bw, const intptr_t bh,
const intptr_t iw, const intptr_t ih,
const intptr_t x, const intptr_t y,
pixel *dst, const ptrdiff_t dst_stride,
const pixel *ref, const ptrdiff_t ref_stride)
{
// find offset in reference of visible block to copy
ref += iclip(y, 0, ih - 1) * PXSTRIDE(ref_stride) + iclip(x, 0, iw - 1);
......
......@@ -528,10 +528,10 @@ static int mc(Dav1dTileContext *const t,
dx + bw4 * h_mul + !!mx * 4 > w ||
dy + bh4 * v_mul + !!my * 4 > h)
{
f->dsp->mc.emu_edge(t->emu_edge, 160 * sizeof(pixel),
refp->p.data[pl], ref_stride,
bw4 * h_mul + !!mx * 7, bh4 * v_mul + !!my * 7,
w, h, dx - !!mx * 3, dy - !!my * 3);
f->dsp->mc.emu_edge(bw4 * h_mul + !!mx * 7, bh4 * v_mul + !!my * 7,
w, h, dx - !!mx * 3, dy - !!my * 3,
t->emu_edge, 160 * sizeof(pixel),
refp->p.data[pl], ref_stride);
ref = &t->emu_edge[160 * !!my * 3 + !!mx * 3];
ref_stride = 160 * sizeof(pixel);
} else {
......@@ -667,9 +667,9 @@ static int warp_affine(Dav1dTileContext *const t,
return -1;
}
if (dx < 3 || dx + 8 + 4 > width || dy < 3 || dy + 8 + 4 > height) {
f->dsp->mc.emu_edge(t->emu_edge, 160 * sizeof(pixel),
refp->p.data[pl], ref_stride,
15, 15, width, height, dx - 3, dy - 3);
f->dsp->mc.emu_edge(15, 15, width, height, dx - 3, dy - 3,
t->emu_edge, 160 * sizeof(pixel),
refp->p.data[pl], ref_stride);
ref_ptr = &t->emu_edge[160 * 3 + 3];
ref_stride = 160 * sizeof(pixel);
} else {
......
......@@ -346,16 +346,17 @@ static void random_offset_for_edge(int *const x, int *const y,
}
static void check_emuedge(Dav1dMCDSPContext *const c) {
ALIGN_STK_32(pixel, c_dst, 135 * 160,);
ALIGN_STK_32(pixel, a_dst, 135 * 160,);
ALIGN_STK_32(pixel, c_dst, 135 * 192,);
ALIGN_STK_32(pixel, a_dst, 135 * 192,);
ALIGN_STK_32(pixel, src, 160 * 160,);
for (int i = 0; i < 160 * 160; i++)
src[i] = rand() & ((1U << BITDEPTH) - 1);
declare_func(void, pixel *dst, ptrdiff_t dst_stride,
const pixel *src, ptrdiff_t src_stride,
int bw, int bh, int iw, int ih, int x, int y);
declare_func(void, intptr_t bw, intptr_t bh, intptr_t iw, intptr_t ih,
intptr_t x, intptr_t y,
pixel *dst, ptrdiff_t dst_stride,
const pixel *src, ptrdiff_t src_stride);
int x, y, iw, ih;
for (int w = 4; w <= 128; w <<= 1)
......@@ -366,18 +367,18 @@ static void check_emuedge(Dav1dMCDSPContext *const c) {
const int bw = w + (rand() & 7);
const int bh = h + (rand() & 7);
random_offset_for_edge(&x, &y, bw, bh, &iw, &ih, edge);
call_ref(c_dst, 160 * sizeof(pixel), src, 160 * sizeof(pixel),
bw, bh, iw, ih, x, y);
call_new(a_dst, 160 * sizeof(pixel), src, 160 * sizeof(pixel),
bw, bh, iw, ih, x, y);
const int res = cmp2d(c_dst, a_dst, 160 * sizeof(pixel), bw, bh);
call_ref(bw, bh, iw, ih, x, y,
c_dst, 192 * sizeof(pixel), src, 160 * sizeof(pixel));
call_new(bw, bh, iw, ih, x, y,
a_dst, 192 * sizeof(pixel), src, 160 * sizeof(pixel));
const int res = cmp2d(c_dst, a_dst, 192 * sizeof(pixel), bw, bh);
if (res != -1) fail();
}
}
for (enum EdgeFlags edge = 1; edge < 0xf; edge <<= 1) {
random_offset_for_edge(&x, &y, w + 7, w + 7, &iw, &ih, edge);
bench_new(a_dst, 160 * sizeof(pixel), src, 160 * sizeof(pixel),
w + 7, w + 7, iw, ih, x, y);
bench_new(w + 7, w + 7, iw, ih, x, y,
a_dst, 192 * sizeof(pixel), src, 160 * sizeof(pixel));
}
}
report("emu_edge");
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment