xf86-video-intel: 3 commits - src/sna/gen6_render.c src/sna/kgem.c src/sna/kgem_debug_gen6.c src/sna/sna_composite.c src/sna/sna.h src/sna/sna_trapezoids.c
Chris Wilson
ickle at kemper.freedesktop.org
Sat Feb 4 12:19:24 PST 2012
src/sna/gen6_render.c | 137 ++++++++++++++++++++++++++++++++++------------
src/sna/kgem.c | 2
src/sna/kgem_debug_gen6.c | 14 +---
src/sna/sna.h | 1
src/sna/sna_composite.c | 25 ++++++--
src/sna/sna_trapezoids.c | 21 +++++--
6 files changed, 145 insertions(+), 55 deletions(-)
New commits:
commit a8ed1a02ada1f8d2f910dfefb150d26c840bf9ea
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Sat Feb 4 20:13:07 2012 +0000
sna: Discard the redundant clear of the unbounded area if already clear
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna_trapezoids.c b/src/sna/sna_trapezoids.c
index c2a4190..28c8a67 100644
--- a/src/sna/sna_trapezoids.c
+++ b/src/sna/sna_trapezoids.c
@@ -2658,6 +2658,13 @@ choose_span(PicturePtr dst,
}
static bool
+sna_drawable_is_clear(DrawablePtr d)
+{
+ struct sna_pixmap *priv = sna_pixmap(get_drawable_pixmap(d));
+ return priv && priv->clear && priv->clear_color == 0;
+}
+
+static bool
mono_trapezoids_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
INT16 src_x, INT16 src_y,
int ntrap, xTrapezoid *traps)
@@ -2666,6 +2673,7 @@ mono_trapezoids_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
BoxRec extents;
int16_t dst_x, dst_y;
int16_t dx, dy;
+ bool was_clear;
int n;
if (NO_SCAN_CONVERTER)
@@ -2709,6 +2717,8 @@ mono_trapezoids_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
if (!mono_init(&mono, 2*ntrap))
return false;
+ was_clear = sna_drawable_is_clear(dst->pDrawable);
+
for (n = 0; n < ntrap; n++) {
if (!xTrapezoidValid(&traps[n]))
continue;
@@ -2741,7 +2751,7 @@ mono_trapezoids_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
mono.op.done(mono.sna, &mono.op);
mono_fini(&mono);
- if (!operator_is_bounded(op)) {
+ if (!was_clear && !operator_is_bounded(op)) {
xPointFixed p1, p2;
if (!mono_init(&mono, 2+2*ntrap))
@@ -4223,6 +4233,7 @@ mono_triangles_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
BoxRec extents;
int16_t dst_x, dst_y;
int16_t dx, dy;
+ bool was_clear;
int n;
mono.sna = to_sna_from_drawable(dst->pDrawable);
@@ -4261,6 +4272,8 @@ mono_triangles_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
src_x + mono.clip.extents.x1 - dst_x - dx,
src_y + mono.clip.extents.y1 - dst_y - dy));
+ was_clear = sna_drawable_is_clear(dst->pDrawable);
+
if (mono_init(&mono, 3*count))
return false;
@@ -4289,7 +4302,7 @@ mono_triangles_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
mono.op.done(mono.sna, &mono.op);
}
- if (!operator_is_bounded(op)) {
+ if (!was_clear && !operator_is_bounded(op)) {
xPointFixed p1, p2;
if (!mono_init(&mono, 2+3*count))
commit b899a4b69696141ca8b897a7abf52649b09f7b3b
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Sat Feb 4 20:07:49 2012 +0000
sna: Always pass the clear colour for PictOpClear
Having made that optimisation for Composite, and then made the
assumption that it is always true in the backends, we failed to clear
the unbounded area outside of a trapezoid since we passed in the
original colour and the operation was optimised as a continuation.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna.h b/src/sna/sna.h
index 9df8cfd..73e2490 100644
--- a/src/sna/sna.h
+++ b/src/sna/sna.h
@@ -274,6 +274,7 @@ struct sna {
void *WakeupData;
CloseScreenProcPtr CloseScreen;
+ PicturePtr clear;
struct {
uint32_t fill_bo;
uint32_t fill_pixel;
diff --git a/src/sna/sna_composite.c b/src/sna/sna_composite.c
index 8eb0c94..5b81596 100644
--- a/src/sna/sna_composite.c
+++ b/src/sna/sna_composite.c
@@ -43,20 +43,18 @@
#define BOUND(v) (INT16) ((v) < MINSHORT ? MINSHORT : (v) > MAXSHORT ? MAXSHORT : (v))
-static PicturePtr clear;
-
Bool sna_composite_create(struct sna *sna)
{
xRenderColor color ={ 0 };
int error;
- clear = CreateSolidPicture(0, &color, &error);
- return clear != NULL;
+ sna->clear = CreateSolidPicture(0, &color, &error);
+ return sna->clear != NULL;
}
void sna_composite_close(struct sna *sna)
{
- FreePicture(clear, 0);
+ FreePicture(sna->clear, 0);
}
static inline bool
@@ -436,7 +434,7 @@ sna_composite(CARD8 op,
if (op == PictOpClear) {
DBG(("%s: discarding source and mask for clear\n", __FUNCTION__));
mask = NULL;
- src = clear;
+ src = sna->clear;
}
if (mask && sna_composite_mask_is_opaque(mask)) {
diff --git a/src/sna/sna_trapezoids.c b/src/sna/sna_trapezoids.c
index eb6c968..c2a4190 100644
--- a/src/sna/sna_trapezoids.c
+++ b/src/sna/sna_trapezoids.c
@@ -2776,7 +2776,7 @@ mono_trapezoids_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
memset(&mono.op, 0, sizeof(mono.op));
if (mono.sna->render.composite(mono.sna,
PictOpClear,
- src, NULL, dst,
+ mono.sna->clear, NULL, dst,
0, 0,
0, 0,
mono.clip.extents.x1, mono.clip.extents.y1,
@@ -4321,7 +4321,7 @@ mono_triangles_span_converter(CARD8 op, PicturePtr src, PicturePtr dst,
memset(&mono.op, 0, sizeof(mono.op));
if (mono.sna->render.composite(mono.sna,
PictOpClear,
- src, NULL, dst,
+ mono.sna->clear, NULL, dst,
0, 0,
0, 0,
mono.clip.extents.x1, mono.clip.extents.y1,
commit c107b90a44abb45c837ff8924939872be5b490eb
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Sat Feb 4 16:33:34 2012 +0000
sna/gen6: Reduce PictOpClear to PictOpSrc (with blending disabled)
The advantage of PictOpSrc is that it writes its results directly to
memory bypassing the blend unit.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/gen6_render.c b/src/sna/gen6_render.c
index 08f9668..d73fda8 100644
--- a/src/sna/gen6_render.c
+++ b/src/sna/gen6_render.c
@@ -579,17 +579,6 @@ gen6_emit_cc(struct sna *sna,
if (render->blend == blend)
return op <= PictOpSrc;
- if (op == PictOpClear) {
- uint32_t src;
-
- /* We can emulate a clear using src, which is beneficial if
- * the blend unit is already disabled.
- */
- src = BLEND_OFFSET(GEN6_BLENDFACTOR_ONE, GEN6_BLENDFACTOR_ZERO);
- if (render->blend == src)
- return true;
- }
-
OUT_BATCH(GEN6_3DSTATE_CC_STATE_POINTERS | (4 - 2));
OUT_BATCH((render->cc_blend + blend) | 1);
if (render->blend == (unsigned)-1) {
@@ -716,8 +705,8 @@ gen6_emit_drawing_rectangle(struct sna *sna,
assert(!too_large(op->dst.x, op->dst.y));
assert(!too_large(op->dst.width, op->dst.height));
- if (sna->render_state.gen6.drawrect_limit == limit &&
- sna->render_state.gen6.drawrect_offset == offset)
+ if (sna->render_state.gen6.drawrect_limit == limit &&
+ sna->render_state.gen6.drawrect_offset == offset)
return false;
/* [DevSNB-C+{W/A}] Before any depth stall flush (including those
@@ -932,6 +921,8 @@ static int gen6_vertex_finish(struct sna *sna)
struct kgem_bo *bo;
unsigned int i;
+ DBG(("%s: used=%d / %d\n", __FUNCTION__,
+ sna->render.vertex_used, sna->render.vertex_size));
assert(sna->render.vertex_used);
/* Note: we only need dword alignment (currently) */
@@ -978,6 +969,10 @@ static int gen6_vertex_finish(struct sna *sna)
kgem_bo_sync__cpu(&sna->kgem, sna->render.vbo);
if (sna->render.vertex_used) {
+ DBG(("%s: copying initial buffer x %d to handle=%d\n",
+ __FUNCTION__,
+ sna->render.vertex_used,
+ sna->render.vbo->handle));
memcpy(sna->render.vertices,
sna->render.vertex_data,
sizeof(float)*sna->render.vertex_used);
@@ -1003,6 +998,8 @@ static void gen6_vertex_close(struct sna *sna)
bo = sna->render.vbo;
if (bo == NULL) {
+ assert(sna->render.vertices == sna->render.vertex_data);
+ assert(sna->render.vertex_used < ARRAY_SIZE(sna->render.vertex_data));
if (sna->kgem.nbatch + sna->render.vertex_used <= sna->kgem.surface) {
DBG(("%s: copy to batch: %d @ %d\n", __FUNCTION__,
sna->render.vertex_used, sna->kgem.nbatch));
@@ -1230,24 +1227,24 @@ gen6_emit_composite_primitive_solid(struct sna *sna,
float f;
} dst;
+ DBG(("%s: [%d+9] = (%d, %d)x(%d, %d)\n", __FUNCTION__,
+ sna->render.vertex_used, r->dst.x, r->dst.y, r->width, r->height));
+
v = sna->render.vertices + sna->render.vertex_used;
sna->render.vertex_used += 9;
+ assert(sna->render.vertex_used <= sna->render.vertex_size);
+ assert(!too_large(r->dst.x + r->width, r->dst.y + r->height));
dst.p.x = r->dst.x + r->width;
dst.p.y = r->dst.y + r->height;
v[0] = dst.f;
- v[1] = 1.;
- v[2] = 1.;
-
dst.p.x = r->dst.x;
v[3] = dst.f;
- v[4] = 0.;
- v[5] = 1.;
-
dst.p.y = r->dst.y;
v[6] = dst.f;
- v[7] = 0.;
- v[8] = 0.;
+
+ v[5] = v[2] = v[1] = 1.;
+ v[8] = v[7] = v[4] = 0.;
}
fastcall static void
@@ -1280,6 +1277,45 @@ gen6_emit_composite_primitive_identity_source(struct sna *sna,
}
fastcall static void
+gen6_emit_composite_primitive_simple_source(struct sna *sna,
+ const struct sna_composite_op *op,
+ const struct sna_composite_rectangles *r)
+{
+ float *v;
+ union {
+ struct sna_coordinate p;
+ float f;
+ } dst;
+
+ float xx = op->src.transform->matrix[0][0];
+ float x0 = op->src.transform->matrix[0][2];
+ float yy = op->src.transform->matrix[1][1];
+ float y0 = op->src.transform->matrix[1][2];
+ float sx = op->src.scale[0];
+ float sy = op->src.scale[1];
+ int16_t tx = op->src.offset[0];
+ int16_t ty = op->src.offset[1];
+
+ v = sna->render.vertices + sna->render.vertex_used;
+ sna->render.vertex_used += 3*3;
+
+ dst.p.x = r->dst.x + r->width;
+ dst.p.y = r->dst.y + r->height;
+ v[0] = dst.f;
+ v[1] = ((r->src.x + r->width + tx) * xx + x0) * sx;
+ v[5] = v[2] = ((r->src.y + r->height + ty) * yy + y0) * sy;
+
+ dst.p.x = r->dst.x;
+ v[3] = dst.f;
+ v[7] = v[4] = ((r->src.x + tx) * xx + x0) * sx;
+
+ dst.p.y = r->dst.y;
+ v[6] = dst.f;
+ v[8] = ((r->src.y + ty) * yy + y0) * sy;
+}
+
+
+fastcall static void
gen6_emit_composite_primitive_affine_source(struct sna *sna,
const struct sna_composite_op *op,
const struct sna_composite_rectangles *r)
@@ -1525,6 +1561,10 @@ static void gen6_emit_vertex_buffer(struct sna *sna,
static void gen6_emit_primitive(struct sna *sna)
{
if (sna->kgem.nbatch == sna->render_state.gen6.last_primitive) {
+ DBG(("%s: continuing previous primitive, start=%d, index=%d\n",
+ __FUNCTION__,
+ sna->render.vertex_start,
+ sna->render.vertex_index));
sna->render_state.gen6.vertex_offset = sna->kgem.nbatch - 5;
return;
}
@@ -1541,6 +1581,8 @@ static void gen6_emit_primitive(struct sna *sna)
OUT_BATCH(0); /* start instance location */
OUT_BATCH(0); /* index buffer offset, ignored */
sna->render.vertex_start = sna->render.vertex_index;
+ DBG(("%s: started new primitive: index=%d\n",
+ __FUNCTION__, sna->render.vertex_start));
sna->render_state.gen6.last_primitive = sna->kgem.nbatch;
}
@@ -1603,6 +1645,7 @@ inline static int gen6_get_rectangles(struct sna *sna,
if (want > 1 && want * op->floats_per_rect > rem)
want = rem / op->floats_per_rect;
+ assert(want > 0);
sna->render.vertex_index += 3*want;
return want;
}
@@ -2156,6 +2199,8 @@ static void gen6_composite_channel_convert(struct sna_composite_channel *channel
static void gen6_render_composite_done(struct sna *sna,
const struct sna_composite_op *op)
{
+ DBG(("%s\n", __FUNCTION__));
+
if (sna->render_state.gen6.vertex_offset) {
gen6_vertex_flush(sna);
gen6_magic_ca_pass(sna, op);
@@ -2488,6 +2533,8 @@ gen6_render_composite(struct sna *sna,
width, height,
tmp);
+ if (op == PictOpClear)
+ op = PictOpSrc;
tmp->op = op;
if (!gen6_composite_set_target(sna, tmp, dst))
return FALSE;
@@ -2585,12 +2632,28 @@ gen6_render_composite(struct sna *sna,
tmp->floats_per_vertex = 5 + 2 * !tmp->is_affine;
} else {
- if (tmp->src.is_solid)
+ if (tmp->src.is_solid) {
+ DBG(("%s: choosing gen6_emit_composite_primitive_solid\n",
+ __FUNCTION__));
tmp->prim_emit = gen6_emit_composite_primitive_solid;
- else if (tmp->src.transform == NULL)
+ } else if (tmp->src.transform == NULL) {
+ DBG(("%s: choosing gen6_emit_composite_primitive_identity_source\n",
+ __FUNCTION__));
tmp->prim_emit = gen6_emit_composite_primitive_identity_source;
- else if (tmp->src.is_affine)
- tmp->prim_emit = gen6_emit_composite_primitive_affine_source;
+ } else if (tmp->src.is_affine) {
+ if (tmp->src.transform->matrix[0][1] == 0 &&
+ tmp->src.transform->matrix[1][0] == 0) {
+ tmp->src.scale[0] /= tmp->src.transform->matrix[2][2];
+ tmp->src.scale[1] /= tmp->src.transform->matrix[2][2];
+ DBG(("%s: choosing gen6_emit_composite_primitive_simple_source\n",
+ __FUNCTION__));
+ tmp->prim_emit = gen6_emit_composite_primitive_simple_source;
+ } else {
+ DBG(("%s: choosing gen6_emit_composite_primitive_affine_source\n",
+ __FUNCTION__));
+ tmp->prim_emit = gen6_emit_composite_primitive_affine_source;
+ }
+ }
tmp->floats_per_vertex = 3 + !tmp->is_affine;
}
@@ -2923,11 +2986,11 @@ fastcall static void
gen6_render_composite_spans_done(struct sna *sna,
const struct sna_composite_spans_op *op)
{
+ DBG(("%s()\n", __FUNCTION__));
+
if (sna->render_state.gen6.vertex_offset)
gen6_vertex_flush(sna);
- DBG(("%s()\n", __FUNCTION__));
-
if (op->base.src.bo)
kgem_bo_destroy(&sna->kgem, op->base.src.bo);
@@ -3193,8 +3256,7 @@ fallback_blt:
if (!gen6_check_format(tmp.src.pict_format))
goto fallback_blt;
- tmp.op = alu == GXcopy ? PictOpSrc : PictOpClear;
-
+ tmp.op = PictOpSrc;
tmp.dst.pixmap = dst;
tmp.dst.width = dst->drawable.width;
tmp.dst.height = dst->drawable.height;
@@ -3373,6 +3435,8 @@ gen6_render_copy_blt(struct sna *sna,
static void
gen6_render_copy_done(struct sna *sna, const struct sna_copy_op *op)
{
+ DBG(("%s()\n", __FUNCTION__));
+
if (sna->render_state.gen6.vertex_offset)
gen6_vertex_flush(sna);
}
@@ -3428,7 +3492,7 @@ fallback:
if (!gen6_check_format(op->base.src.pict_format))
goto fallback;
- op->base.op = alu == GXcopy ? PictOpSrc : PictOpClear;
+ op->base.op = PictOpSrc;
op->base.dst.pixmap = dst;
op->base.dst.width = dst->drawable.width;
@@ -3574,9 +3638,10 @@ gen6_render_fill_boxes(struct sna *sna,
return FALSE;
#endif
- if (op == PictOpClear)
+ if (op == PictOpClear) {
pixel = 0;
- else if (!sna_get_pixel_from_rgba(&pixel,
+ op = PictOpSrc;
+ } else if (!sna_get_pixel_from_rgba(&pixel,
color->red,
color->green,
color->blue,
@@ -3744,6 +3809,8 @@ gen6_render_op_fill_boxes(struct sna *sna,
static void
gen6_render_op_fill_done(struct sna *sna, const struct sna_fill_op *op)
{
+ DBG(("%s()\n", __FUNCTION__));
+
if (sna->render_state.gen6.vertex_offset)
gen6_vertex_flush(sna);
kgem_bo_destroy(&sna->kgem, op->base.src.bo);
@@ -3781,7 +3848,7 @@ gen6_render_fill(struct sna *sna, uint8_t alu,
if (alu == GXclear)
color = 0;
- op->base.op = color == 0 ? PictOpClear : PictOpSrc;
+ op->base.op = PictOpSrc;
op->base.dst.pixmap = dst;
op->base.dst.width = dst->drawable.width;
@@ -3874,7 +3941,7 @@ gen6_render_fill_one(struct sna *sna, PixmapPtr dst, struct kgem_bo *bo,
if (alu == GXclear)
color = 0;
- tmp.op = color == 0 ? PictOpClear : PictOpSrc;
+ tmp.op = PictOpSrc;
tmp.dst.pixmap = dst;
tmp.dst.width = dst->drawable.width;
@@ -3976,7 +4043,7 @@ gen6_render_clear(struct sna *sna, PixmapPtr dst, struct kgem_bo *bo)
if (too_large(dst->drawable.width, dst->drawable.height))
return gen6_render_clear_try_blt(sna, dst, bo);
- tmp.op = PictOpClear;
+ tmp.op = PictOpSrc;
tmp.dst.pixmap = dst;
tmp.dst.width = dst->drawable.width;
diff --git a/src/sna/kgem.c b/src/sna/kgem.c
index 757bad1..259666e 100644
--- a/src/sna/kgem.c
+++ b/src/sna/kgem.c
@@ -2917,7 +2917,7 @@ void *kgem_bo_map(struct kgem *kgem, struct kgem_bo *bo)
void *kgem_bo_map__debug(struct kgem *kgem, struct kgem_bo *bo)
{
if (bo->map)
- return bo->map;
+ return CPU_MAP(bo->map);
kgem_trim_vma_cache(kgem, MAP_GTT, bucket(bo));
return bo->map = gem_mmap(kgem->fd, bo->handle, bytes(bo),
diff --git a/src/sna/kgem_debug_gen6.c b/src/sna/kgem_debug_gen6.c
index f748da6..72ed299 100644
--- a/src/sna/kgem_debug_gen6.c
+++ b/src/sna/kgem_debug_gen6.c
@@ -42,7 +42,6 @@
static struct state {
struct vertex_buffer {
int handle;
- void *base;
const char *ptr;
int pitch;
@@ -67,7 +66,7 @@ static void gen6_update_vertex_buffer(struct kgem *kgem, const uint32_t *data)
{
uint32_t reloc = sizeof(uint32_t) * (&data[1] - kgem->batch);
struct kgem_bo *bo = NULL;
- void *base, *ptr;
+ void *base;
int i;
for (i = 0; i < kgem->nreloc; i++)
@@ -85,13 +84,12 @@ static void gen6_update_vertex_buffer(struct kgem *kgem, const uint32_t *data)
assert(&bo->request != &kgem->next_request->buffers);
base = kgem_bo_map__debug(kgem, bo);
}
- ptr = (char *)base + kgem->reloc[i].delta;
+ base = (char *)base + kgem->reloc[i].delta;
i = data[0] >> 26;
state.vb[i].current = bo;
- state.vb[i].base = base;
- state.vb[i].ptr = ptr;
+ state.vb[i].ptr = base;
state.vb[i].pitch = data[0] & 0x7ff;
}
@@ -268,10 +266,8 @@ static void indirect_vertex_out(struct kgem *kgem, uint32_t v)
const struct vertex_buffer *vb = &state.vb[ve->buffer];
const void *ptr = vb->ptr + v * vb->pitch + ve->offset;
- if (!ve->valid)
- continue;
-
- ve_out(ve, ptr);
+ if (ve->valid)
+ ve_out(ve, ptr);
while (++i <= state.num_ve && !state.ve[i].valid)
;
diff --git a/src/sna/sna_composite.c b/src/sna/sna_composite.c
index 0faad84..8eb0c94 100644
--- a/src/sna/sna_composite.c
+++ b/src/sna/sna_composite.c
@@ -792,6 +792,21 @@ sna_composite_rectangles(CARD8 op,
}
} else
sna_damage_add(&priv->gpu_damage, ®ion);
+ } else if (op <= PictOpSrc &&
+ region.data == NULL &&
+ region.extents.x2 - region.extents.x1 == pixmap->drawable.width &&
+ region.extents.y2 - region.extents.y1 == pixmap->drawable.height) {
+ priv->clear = true;
+ priv->clear_color = 0;
+ if (op == PictOpSrc)
+ sna_get_pixel_from_rgba(&priv->clear_color,
+ color->red,
+ color->green,
+ color->blue,
+ color->alpha,
+ dst->format);
+ DBG(("%s: marking clear [%08x]\n",
+ __FUNCTION__, priv->clear_color));
}
goto done;
More information about the xorg-commit
mailing list