Aurelien
/
xf86-video-intel
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

sna: Determine whether to use a partial proxy based on the pitch 

On gen4+ devices the maximum render pitch is much larger than is simply
required for the maximum coordinates. This makes it possible to use
proxy textures as a subimage into the oversized texture without having
to blit into a temporary copy for virtually every single bo we use.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This commit is contained in:
Chris Wilson 2012-01-29 11:02:38 +00:00
parent 65466f8626
commit 5c6255ba2f
10 changed files with 173 additions and 61 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
src/sna/gen2_render.c
View File

@ -55,6 +55,7 @@
#define PREFER_BLT_COPY 1
#define MAX_3D_SIZE 2048
#define MAX_3D_PITCH 8192
#define BATCH(v) batch_emit(sna, v)
#define BATCH_F(v) batch_emit_float(sna, v)
@ -547,7 +548,7 @@ gen2_get_batch(struct sna *sna)
static void gen2_emit_target(struct sna *sna, const struct sna_composite_op *op)
{
assert(op->dst.bo->pitch >= 8 && op->dst.bo->pitch <= 8192);
assert(op->dst.bo->pitch >= 8 && op->dst.bo->pitch <= MAX_3D_PITCH);
assert(sna->render_state.gen2.vertex_offset == 0);
if (sna->render_state.gen2.target == op->dst.bo->unique_id) {
@ -1736,7 +1737,7 @@ gen2_render_composite(struct sna *sna,
tmp->op = op;
if (too_large(tmp->dst.width, tmp->dst.height) ||
tmp->dst.bo->pitch > 8192) {
tmp->dst.bo->pitch > MAX_3D_PITCH) {
if (!sna_render_composite_redirect(sna, tmp,
dst_x, dst_y, width, height))
return FALSE;
@ -2192,7 +2193,7 @@ gen2_render_composite_spans(struct sna *sna,
tmp->base.op = op;
if (too_large(tmp->base.dst.width, tmp->base.dst.height) ||
tmp->base.dst.bo->pitch > 8192) {
tmp->base.dst.bo->pitch > MAX_3D_PITCH) {
if (!sna_render_composite_redirect(sna, &tmp->base,
dst_x, dst_y, width, height))
return FALSE;
@ -2388,7 +2389,7 @@ gen2_render_fill_boxes(struct sna *sna,
color->red, color->green, color->blue, color->alpha));
if (too_large(dst->drawable.width, dst->drawable.height) ||
dst_bo->pitch < 8 || dst_bo->pitch > 8192 ||
dst_bo->pitch < 8 || dst_bo->pitch > MAX_3D_PITCH ||
!gen2_check_dst_format(format)) {
DBG(("%s: try blt, too large or incompatible destination\n",
__FUNCTION__));
@ -2589,7 +2590,7 @@ gen2_render_fill(struct sna *sna, uint8_t alu,
/* Must use the BLT if we can't RENDER... */
if (too_large(dst->drawable.width, dst->drawable.height) ||
dst_bo->pitch < 8 || dst_bo->pitch > 8192)
dst_bo->pitch < 8 || dst_bo->pitch > MAX_3D_PITCH)
return sna_blt_fill(sna, alu,
dst_bo, dst->drawable.bitsPerPixel,
color,
@ -2665,7 +2666,7 @@ gen2_render_fill_one(struct sna *sna, PixmapPtr dst, struct kgem_bo *bo,
/* Must use the BLT if we can't RENDER... */
if (too_large(dst->drawable.width, dst->drawable.height) ||
bo->pitch < 8 || bo->pitch > 8192)
bo->pitch < 8 || bo->pitch > MAX_3D_PITCH)
return gen2_render_fill_one_try_blt(sna, dst, bo, color,
x1, y1, x2, y2, alu);
@ -2832,9 +2833,9 @@ gen2_render_copy_boxes(struct sna *sna, uint8_t alu,
if (src_bo == dst_bo || /* XXX handle overlap using 3D ? */
too_large(src->drawable.width, src->drawable.height) ||
src_bo->pitch > 8192 ||
src_bo->pitch > MAX_3D_PITCH ||
too_large(dst->drawable.width, dst->drawable.height) ||
dst_bo->pitch < 8 || dst_bo->pitch > 8192)
dst_bo->pitch < 8 || dst_bo->pitch > MAX_3D_PITCH)
return sna_blt_copy_boxes_fallback(sna, alu,
src, src_bo, src_dx, src_dy,
dst, dst_bo, dst_dx, dst_dy,
@ -2957,7 +2958,8 @@ gen2_render_copy(struct sna *sna, uint8_t alu,
/* Must use the BLT if we can't RENDER... */
if (too_large(src->drawable.width, src->drawable.height) ||
too_large(dst->drawable.width, dst->drawable.height) ||
src_bo->pitch > 8192 || dst_bo->pitch < 8 || dst_bo->pitch > 8192) {
src_bo->pitch > MAX_3D_PITCH ||
dst_bo->pitch < 8 || dst_bo->pitch > MAX_3D_PITCH) {
if (!sna_blt_compare_depth(&src->drawable, &dst->drawable))
return FALSE;
@ -3045,5 +3047,6 @@ Bool gen2_render_init(struct sna *sna)
render->flush = gen2_render_flush;
render->max_3d_size = MAX_3D_SIZE;
render->max_3d_pitch = MAX_3D_PITCH;
return TRUE;
}

16
src/sna/gen3_render.c
View File

@ -65,6 +65,7 @@ enum {
};
#define MAX_3D_SIZE 2048
#define MAX_3D_PITCH 8192
#define OUT_BATCH(v) batch_emit(sna, v)
#define OUT_BATCH_F(v) batch_emit_float(sna, v)
@ -143,7 +144,7 @@ static inline uint32_t gen3_buf_tiling(uint32_t tiling)
static inline Bool
gen3_check_pitch_3d(struct kgem_bo *bo)
{
return bo->pitch <= 8192;
return bo->pitch <= MAX_3D_PITCH;
}
static uint32_t gen3_get_blend_cntl(int op,
@ -3826,9 +3827,9 @@ gen3_render_copy_boxes(struct sna *sna, uint8_t alu,
if (!(alu == GXcopy || alu == GXclear) ||
src_bo == dst_bo || /* XXX handle overlap using 3D ? */
src_bo->pitch > 8192 ||
src_bo->pitch > MAX_3D_PITCH ||
too_large(src->drawable.width, src->drawable.height) ||
dst_bo->pitch > 8192 ||
dst_bo->pitch > MAX_3D_PITCH ||
too_large(dst->drawable.width, dst->drawable.height))
return sna_blt_copy_boxes_fallback(sna, alu,
src, src_bo, src_dx, src_dy,
@ -3959,7 +3960,7 @@ gen3_render_copy(struct sna *sna, uint8_t alu,
if (!(alu == GXcopy || alu == GXclear) ||
too_large(src->drawable.width, src->drawable.height) ||
too_large(dst->drawable.width, dst->drawable.height) ||
src_bo->pitch > 8192 || dst_bo->pitch > 8192) {
src_bo->pitch > MAX_3D_PITCH || dst_bo->pitch > MAX_3D_PITCH) {
if (!sna_blt_compare_depth(&src->drawable, &dst->drawable))
return FALSE;
@ -4083,7 +4084,7 @@ gen3_render_fill_boxes(struct sna *sna,
color->red, color->green, color->blue, color->alpha));
if (too_large(dst->drawable.width, dst->drawable.height) ||
dst_bo->pitch > 8192 ||
dst_bo->pitch > MAX_3D_PITCH ||
!gen3_check_dst_format(format)) {
DBG(("%s: try blt, too large or incompatible destination\n",
__FUNCTION__));
@ -4265,7 +4266,7 @@ gen3_render_fill(struct sna *sna, uint8_t alu,
/* Must use the BLT if we can't RENDER... */
if (!(alu == GXcopy || alu == GXclear) ||
too_large(dst->drawable.width, dst->drawable.height) ||
dst_bo->pitch > 8192)
dst_bo->pitch > MAX_3D_PITCH)
return sna_blt_fill(sna, alu,
dst_bo, dst->drawable.bitsPerPixel,
color,
@ -4346,7 +4347,7 @@ gen3_render_fill_one(struct sna *sna, PixmapPtr dst, struct kgem_bo *bo,
/* Must use the BLT if we can't RENDER... */
if (!(alu == GXcopy || alu == GXclear) ||
too_large(dst->drawable.width, dst->drawable.height) ||
bo->pitch > 8192)
bo->pitch > MAX_3D_PITCH)
return gen3_render_fill_one_try_blt(sna, dst, bo, color,
x1, y1, x2, y2, alu);
@ -4424,5 +4425,6 @@ Bool gen3_render_init(struct sna *sna)
render->fini = gen3_render_fini;
render->max_3d_size = MAX_3D_SIZE;
render->max_3d_pitch = MAX_3D_PITCH;
return TRUE;
}

1
src/sna/gen4_render.c
View File

@ -3220,5 +3220,6 @@ Bool gen4_render_init(struct sna *sna)
sna->render.fini = gen4_render_fini;
sna->render.max_3d_size = 8192;
sna->render.max_3d_pitch = 1 << 18;
return TRUE;
}

1
src/sna/gen5_render.c
View File

@ -3702,5 +3702,6 @@ Bool gen5_render_init(struct sna *sna)
sna->render.fini = gen5_render_fini;
sna->render.max_3d_size = MAX_3D_SIZE;
sna->render.max_3d_pitch = 1 << 18;
return TRUE;
}

1
src/sna/gen6_render.c
View File

@ -4044,5 +4044,6 @@ Bool gen6_render_init(struct sna *sna)
sna->render.fini = gen6_render_fini;
sna->render.max_3d_size = GEN6_MAX_SIZE;
sna->render.max_3d_pitch = 1 << 18;
return TRUE;
}

1
src/sna/gen7_render.c
View File

@ -4097,5 +4097,6 @@ Bool gen7_render_init(struct sna *sna)
sna->render.fini = gen7_render_fini;
sna->render.max_3d_size = GEN7_MAX_SIZE;
sna->render.max_3d_pitch = 1 << 18;
return TRUE;
}

33
src/sna/kgem.c
View File

@ -704,6 +704,39 @@ static uint32_t kgem_untiled_pitch(struct kgem *kgem,
return ALIGN(width, scanout ? 64 : kgem->min_alignment);
}
void kgem_get_tile_size(struct kgem *kgem, int tiling,
int *tile_width, int *tile_height, int *tile_size)
{
if (kgem->gen < 30) {
if (tiling) {
*tile_width = 512;
*tile_height = 16;
*tile_size = 2048;
} else {
*tile_width = 1;
*tile_height = 1;
*tile_size = 1;
}
} else switch (tiling) {
default:
case I915_TILING_NONE:
*tile_width = 1;
*tile_height = 1;
*tile_size = 1;
break;
case I915_TILING_X:
*tile_width = 512;
*tile_height = 8;
*tile_size = 4096;
break;
case I915_TILING_Y:
*tile_width = kgem->gen <= 30 ? 512 : 128;
*tile_height = 32;
*tile_size = 4096;
break;
}
}
static uint32_t kgem_surface_size(struct kgem *kgem,
bool relaxed_fencing,
bool scanout,

2
src/sna/kgem.h
View File

@ -361,6 +361,8 @@ Bool kgem_bo_write(struct kgem *kgem, struct kgem_bo *bo,
const void *data, int length);
int kgem_bo_fenced_size(struct kgem *kgem, struct kgem_bo *bo);
void kgem_get_tile_size(struct kgem *kgem, int tiling,
int *tile_width, int *tile_height, int *tile_size);
static inline int kgem_bo_size(struct kgem_bo *bo)
{

157
src/sna/sna_render.c
View File

@ -816,10 +816,27 @@ sna_render_picture_partial(struct sna *sna,
struct kgem_bo *bo = NULL;
PixmapPtr pixmap = get_drawable_pixmap(picture->pDrawable);
BoxRec box;
int tile_width, tile_height, tile_size;
int offset;
DBG(("%s (%d, %d)x(%d, %d) [dst=(%d, %d)]\n",
__FUNCTION__, x, y, w, h, dst_x, dst_y));
if (use_cpu_bo(sna, pixmap, &box)) {
if (!sna_pixmap_move_to_cpu(pixmap, MOVE_READ))
return 0;
bo = sna_pixmap(pixmap)->cpu_bo;
} else {
if (!sna_pixmap_force_to_gpu(pixmap, MOVE_READ))
return 0;
bo = sna_pixmap(pixmap)->gpu_bo;
}
if (bo->pitch > sna->render.max_3d_pitch)
return 0;
box.x1 = x;
box.y1 = y;
box.x2 = x + w;
@ -855,51 +872,65 @@ sna_render_picture_partial(struct sna *sna,
}
}
/* Presume worst case tile-row alignment for Y-tiling */
box.y1 = box.y1 & (64 - 1);
box.y2 = ALIGN(box.y2, 64);
kgem_get_tile_size(&sna->kgem, bo->tiling,
&tile_width, &tile_height, &tile_size);
/* Ensure we align to an even tile row */
box.y1 = box.y1 & ~(2*tile_height - 1);
box.y2 = ALIGN(box.y2, 2*tile_height);
if (box.y2 > pixmap->drawable.height)
box.y2 = pixmap->drawable.height;
box.x1 = box.x1 & ~(tile_width * 8 / pixmap->drawable.bitsPerPixel - 1);
box.x2 = ALIGN(box.x2, tile_width * 8 / pixmap->drawable.bitsPerPixel);
if (box.x2 > pixmap->drawable.width)
box.x2 = pixmap->drawable.width;
w = box.x2 - box.x1;
h = box.y2 - box.y1;
DBG(("%s box=(%d, %d), (%d, %d): (%d, %d)/(%d, %d)\n", __FUNCTION__,
box.x1, box.y1, box.x2, box.y2, w, h,
pixmap->drawable.width, pixmap->drawable.height));
if (w <= 0 || h <= 0 || h > sna->render.max_3d_size)
if (w <= 0 || h <= 0 ||
w > sna->render.max_3d_size ||
h > sna->render.max_3d_size)
return 0;
memset(&channel->embedded_transform,
0,
sizeof(channel->embedded_transform));
channel->embedded_transform.matrix[0][0] = 1 << 16;
channel->embedded_transform.matrix[0][2] = 0;
channel->embedded_transform.matrix[1][1] = 1 << 16;
channel->embedded_transform.matrix[1][2] = -box.y1 << 16;
channel->embedded_transform.matrix[2][2] = 1 << 16;
if (channel->transform)
/* How many tiles across are we? */
offset = box.x1 * pixmap->drawable.bitsPerPixel / 8 / tile_width * tile_size;
channel->bo = kgem_create_proxy(bo,
box.y1 * bo->pitch + offset,
h * bo->pitch);
if (channel->bo == NULL)
return 0;
channel->bo->pitch = bo->pitch;
if (channel->transform) {
memset(&channel->embedded_transform,
0,
sizeof(channel->embedded_transform));
channel->embedded_transform.matrix[0][0] = 1 << 16;
channel->embedded_transform.matrix[0][2] = -box.x1 << 16;
channel->embedded_transform.matrix[1][1] = 1 << 16;
channel->embedded_transform.matrix[1][2] = -box.y1 << 16;
channel->embedded_transform.matrix[2][2] = 1 << 16;
pixman_transform_multiply(&channel->embedded_transform,
&channel->embedded_transform,
channel->transform);
channel->transform = &channel->embedded_transform;
if (use_cpu_bo(sna, pixmap, &box)) {
if (!sna_pixmap_move_to_cpu(pixmap, MOVE_READ))
return 0;
bo = sna_pixmap(pixmap)->cpu_bo;
channel->transform = &channel->embedded_transform;
} else {
if (!sna_pixmap_force_to_gpu(pixmap, MOVE_READ))
return 0;
bo = sna_pixmap(pixmap)->gpu_bo;
x -= box.x1;
y -= box.y1;
}
channel->offset[0] = x - dst_x;
channel->offset[1] = y - dst_y;
channel->scale[0] = 1.f/pixmap->drawable.width;
channel->scale[0] = 1.f/w;
channel->scale[1] = 1.f/h;
channel->width = pixmap->drawable.width;
channel->width = w;
channel->height = h;
channel->bo = kgem_create_proxy(bo, box.y1 * bo->pitch, h * bo->pitch);
return channel->bo != NULL;
return 1;
}
int
@ -927,8 +958,7 @@ sna_render_picture_extract(struct sna *sna,
return -1;
}
if (pixmap->drawable.width < sna->render.max_3d_size &&
sna_render_picture_partial(sna, picture, channel,
if (sna_render_picture_partial(sna, picture, channel,
x, y, w, h,
dst_x, dst_y))
return 1;
@ -1527,30 +1557,67 @@ sna_render_composite_redirect(struct sna *sna,
return FALSE;
}
if (op->dst.pixmap->drawable.width <= sna->render.max_3d_size) {
int y1, y2;
if (op->dst.bo->pitch <= sna->render.max_3d_pitch) {
int tile_width, tile_height, tile_size;
BoxRec box;
int w, h;
assert(op->dst.pixmap->drawable.height > sna->render.max_3d_size);
y1 = y + op->dst.y;
y2 = y1 + height;
y1 &= y1 & (64 - 1);
y2 = ALIGN(y2, 64);
kgem_get_tile_size(&sna->kgem, op->dst.bo->tiling,
&tile_width, &tile_height, &tile_size);
box.x1 = x;
box.x2 = x + width;
box.y1 = y;
box.y2 = y + height;
/* Ensure we align to an even tile row */
box.y1 = box.y1 & ~(2*tile_height - 1);
box.y2 = ALIGN(box.y2, 2*tile_height);
if (box.y2 > op->dst.pixmap->drawable.height)
box.y2 = op->dst.pixmap->drawable.height;
box.x1 = box.x1 & ~(tile_width * 8 / op->dst.pixmap->drawable.bitsPerPixel - 1);
box.x2 = ALIGN(box.x2, tile_width * 8 / op->dst.pixmap->drawable.bitsPerPixel);
if (box.x2 > op->dst.pixmap->drawable.width)
box.x2 = op->dst.pixmap->drawable.width;
w = box.x2 - box.x1;
h = box.y2 - box.y1;
DBG(("%s box=(%d, %d), (%d, %d): (%d, %d)/(%d, %d)\n", __FUNCTION__,
box.x1, box.y1, box.x2, box.y2, w, h,
op->dst.pixmap->drawable.width,
op->dst.pixmap->drawable.height));
if (w <= sna->render.max_3d_size &&
h <= sna->render.max_3d_size) {
int offset;
if (y2 - y1 <= sna->render.max_3d_size) {
t->box.x2 = t->box.x1 = op->dst.x;
t->box.y2 = t->box.y1 = op->dst.y;
t->real_bo = priv->gpu_bo;
t->real_bo = op->dst.bo;
t->real_damage = op->damage;
if (op->damage) {
t->damage = sna_damage_create();
op->damage = &t->damage;
}
op->dst.bo = kgem_create_proxy(priv->gpu_bo,
y1 * priv->gpu_bo->pitch,
(y2 - y1) * priv->gpu_bo->pitch);
op->dst.y += -y1;
op->dst.height = y2 - y1;
/* How many tiles across are we? */
offset = box.x1 * op->dst.pixmap->drawable.bitsPerPixel / 8 / tile_width * tile_size;
op->dst.bo = kgem_create_proxy(op->dst.bo,
box.y1 * op->dst.bo->pitch + offset,
h * op->dst.bo->pitch);
if (!op->dst.bo) {
t->real_bo = NULL;
if (t->damage)
__sna_damage_destroy(t->damage);
return FALSE;
}
op->dst.bo->pitch = t->real_bo->pitch;
op->dst.x += -box.x1;
op->dst.y += -box.y1;
op->dst.width = w;
op->dst.height = h;
return TRUE;
}
}
@ -1583,7 +1650,7 @@ sna_render_composite_redirect(struct sna *sna,
return FALSE;
}
t->real_bo = priv->gpu_bo;
t->real_bo = op->dst.bo;
t->real_damage = op->damage;
if (op->damage) {
t->damage = sna_damage_create();

1
src/sna/sna_render.h
View File

@ -186,6 +186,7 @@ struct sna_copy_op {
struct sna_render {
int max_3d_size;
int max_3d_pitch;
Bool (*composite)(struct sna *sna, uint8_t op,
PicturePtr dst, PicturePtr src, PicturePtr mask,