xf86-video-intel: 3 commits - src/sna/kgem.c src/sna/kgem.h src/sna/sna_dri2.c src/sna/sna_trapezoids.c src/sna/sna_trapezoids.h src/sna/sna_trapezoids_imprecise.c src/sna/sna_trapezoids_mono.c src/sna/sna_trapezoids_precise.c
Chris Wilson
ickle at kemper.freedesktop.org
Tue Oct 21 23:27:17 PDT 2014
src/sna/kgem.c | 381 +++++++++++++++++++++----------------
src/sna/kgem.h | 28 +-
src/sna/sna_dri2.c | 3
src/sna/sna_trapezoids.c | 9
src/sna/sna_trapezoids.h | 37 ++-
src/sna/sna_trapezoids_imprecise.c | 14 -
src/sna/sna_trapezoids_mono.c | 4
src/sna/sna_trapezoids_precise.c | 82 +++++++
8 files changed, 367 insertions(+), 191 deletions(-)
New commits:
commit 6e486e11a53f93b628a3a82d5b086e63b7dc6c1e
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Wed Oct 22 07:24:10 2014 +0100
sna/dri2: Fix interoperation with PRIME and older Xorg
When using the fake triple buffer implementation with recent enough Xorg
to support both real triple buffering and PRIME, we have to disable use
of SwapLimit as that freezes the client when we are instead faking the
triple buffer. (We have to use the fake support on those Xservers to
workaround crashes inside the Xserver.)
Reported-by: Paulo Dias <paulo.miguel.dias at gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=85144#c12
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna_dri2.c b/src/sna/sna_dri2.c
index 6359377..c8c71c5 100644
--- a/src/sna/sna_dri2.c
+++ b/src/sna/sna_dri2.c
@@ -309,6 +309,9 @@ sna_dri2_reuse_buffer(DrawablePtr draw, DRI2BufferPtr buffer)
static bool swap_limit(DrawablePtr draw, int limit)
{
+ if (!xorg_can_triple_buffer())
+ return false;
+
DBG(("%s: draw=%ld setting swap limit to %d\n", __FUNCTION__, (long)draw->id, limit));
DRI2SwapLimit(draw, limit);
return true;
commit de96fbf66cf9a0df85cfd1dad8cf786a1d6f8325
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Tue Oct 21 12:20:29 2014 +0100
sna/trapezoids: Prefer GPU for AddTraps()
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 d9390db..8131cb2 100644
--- a/src/sna/sna_trapezoids.c
+++ b/src/sna/sna_trapezoids.c
@@ -822,11 +822,13 @@ void
sna_add_traps(PicturePtr picture, INT16 x, INT16 y, int n, xTrap *t)
{
struct sna *sna;
+ PixmapPtr pixmap;
DBG(("%s (%d, %d) x %d\n", __FUNCTION__, x, y, n));
- sna = to_sna_from_drawable(picture->pDrawable);
- if (is_gpu(sna, picture->pDrawable, PREFER_GPU_SPANS)) {
+ pixmap = get_drawable_pixmap(picture->pDrawable);
+ sna = to_sna_from_pixmap(pixmap);
+ if (is_gpu_dst(sna_pixmap(pixmap))) {
if (trap_span_converter(sna, picture, x, y, n, t))
return;
}
@@ -840,8 +842,7 @@ sna_add_traps(PicturePtr picture, INT16 x, INT16 y, int n, xTrap *t)
}
DBG(("%s -- fallback\n", __FUNCTION__));
- if (sna_drawable_move_to_cpu(picture->pDrawable,
- MOVE_READ | MOVE_WRITE)) {
+ if (sna_pixmap_move_to_cpu(pixmap, MOVE_READ | MOVE_WRITE)) {
pixman_image_t *image;
int dx, dy;
diff --git a/src/sna/sna_trapezoids.h b/src/sna/sna_trapezoids.h
index c057cab..64c7747 100644
--- a/src/sna/sna_trapezoids.h
+++ b/src/sna/sna_trapezoids.h
@@ -54,12 +54,6 @@ mono_trapezoid_span_inplace(struct sna *sna,
int ntrap, xTrapezoid *traps);
bool
-mono_trap_span_converter(struct sna *sna,
- PicturePtr dst,
- INT16 x, INT16 y,
- int ntrap, xTrap *traps);
-
-bool
mono_triangles_span_converter(struct sna *sna,
CARD8 op, PicturePtr src, PicturePtr dst,
INT16 src_x, INT16 src_y,
@@ -210,10 +204,39 @@ trapezoid_span_fallback(CARD8 op, PicturePtr src, PicturePtr dst,
}
bool
+mono_trap_span_converter(struct sna *sna,
+ PicturePtr dst,
+ INT16 x, INT16 y,
+ int ntrap, xTrap *traps);
+
+bool
+precise_trap_span_converter(struct sna *sna,
+ PicturePtr dst,
+ INT16 src_x, INT16 src_y,
+ int ntrap, xTrap *trap);
+
+bool
+imprecise_trap_span_converter(struct sna *sna,
+ PicturePtr dst,
+ INT16 src_x, INT16 src_y,
+ int ntrap, xTrap *trap);
+
+static inline bool
trap_span_converter(struct sna *sna,
PicturePtr dst,
INT16 src_x, INT16 src_y,
- int ntrap, xTrap *trap);
+ int ntrap, xTrap *trap)
+{
+ if (NO_SCAN_CONVERTER)
+ return false;
+
+ if (dst->polyEdge == PolyEdgeSharp || dst->pDrawable->depth < 8)
+ return mono_trap_span_converter(sna, dst, src_x, src_y, ntrap, trap);
+ else if (dst->polyMode == PolyModePrecise)
+ return precise_trap_span_converter(sna, dst, src_x, src_y, ntrap, trap);
+ else
+ return imprecise_trap_span_converter(sna, dst, src_x, src_y, ntrap, trap);
+}
bool
trap_mask_converter(struct sna *sna,
diff --git a/src/sna/sna_trapezoids_imprecise.c b/src/sna/sna_trapezoids_imprecise.c
index 0887d4f..f88a8ae 100644
--- a/src/sna/sna_trapezoids_imprecise.c
+++ b/src/sna/sna_trapezoids_imprecise.c
@@ -3233,10 +3233,10 @@ imprecise_trapezoid_span_fallback(CARD8 op, PicturePtr src, PicturePtr dst,
}
bool
-trap_span_converter(struct sna *sna,
- PicturePtr dst,
- INT16 src_x, INT16 src_y,
- int ntrap, xTrap *trap)
+imprecise_trap_span_converter(struct sna *sna,
+ PicturePtr dst,
+ INT16 src_x, INT16 src_y,
+ int ntrap, xTrap *trap)
{
struct sna_composite_spans_op tmp;
struct tor tor;
@@ -3244,15 +3244,9 @@ trap_span_converter(struct sna *sna,
pixman_region16_t *clip;
int dx, dy, n;
- if (NO_SCAN_CONVERTER)
- return false;
-
if (dst->pDrawable->depth < 8)
return false;
- if (dst->polyEdge == PolyEdgeSharp)
- return mono_trap_span_converter(sna, dst, src_x, src_y, ntrap, trap);
-
if (!sna->render.check_composite_spans(sna, PictOpAdd, sna->render.white_picture, dst,
dst->pCompositeClip->extents.x2 - dst->pCompositeClip->extents.x1,
dst->pCompositeClip->extents.y2 - dst->pCompositeClip->extents.y1,
diff --git a/src/sna/sna_trapezoids_mono.c b/src/sna/sna_trapezoids_mono.c
index be95875..4ad4573 100644
--- a/src/sna/sna_trapezoids_mono.c
+++ b/src/sna/sna_trapezoids_mono.c
@@ -1286,6 +1286,10 @@ mono_trap_span_converter(struct sna *sna,
mono.clip.extents.x2 - mono.clip.extents.x1,
mono.clip.extents.y2 - mono.clip.extents.y1,
COMPOSITE_PARTIAL, memset(&mono.op, 0, sizeof(mono.op)))) {
+ if (mono.clip.data == NULL && mono.op.damage == NULL)
+ mono.span = mono_span__fast;
+ else
+ mono.span = mono_span;
mono_render(&mono);
mono.op.done(mono.sna, &mono.op);
}
diff --git a/src/sna/sna_trapezoids_precise.c b/src/sna/sna_trapezoids_precise.c
index 2c5dad5..9925654 100644
--- a/src/sna/sna_trapezoids_precise.c
+++ b/src/sna/sna_trapezoids_precise.c
@@ -3538,3 +3538,85 @@ skip:
REGION_UNINIT(NULL, &clip);
return true;
}
+
+bool
+precise_trap_span_converter(struct sna *sna,
+ PicturePtr dst,
+ INT16 src_x, INT16 src_y,
+ int ntrap, xTrap *trap)
+{
+ struct sna_composite_spans_op tmp;
+ struct tor tor;
+ BoxRec extents;
+ pixman_region16_t *clip;
+ int dx, dy, n;
+
+ if (dst->pDrawable->depth < 8)
+ return false;
+
+ if (!sna->render.check_composite_spans(sna, PictOpAdd, sna->render.white_picture, dst,
+ dst->pCompositeClip->extents.x2 - dst->pCompositeClip->extents.x1,
+ dst->pCompositeClip->extents.y2 - dst->pCompositeClip->extents.y1,
+ 0)) {
+ DBG(("%s: fallback -- composite spans not supported\n",
+ __FUNCTION__));
+ return false;
+ }
+
+ clip = dst->pCompositeClip;
+ extents = *RegionExtents(clip);
+ dx = dst->pDrawable->x;
+ dy = dst->pDrawable->y;
+
+ DBG(("%s: after clip -- extents (%d, %d), (%d, %d), delta=(%d, %d)\n",
+ __FUNCTION__,
+ extents.x1, extents.y1,
+ extents.x2, extents.y2,
+ dx, dy));
+
+ memset(&tmp, 0, sizeof(tmp));
+ if (!sna->render.composite_spans(sna, PictOpAdd, sna->render.white_picture, dst,
+ 0, 0,
+ extents.x1, extents.y1,
+ extents.x2 - extents.x1,
+ extents.y2 - extents.y1,
+ 0,
+ &tmp)) {
+ DBG(("%s: fallback -- composite spans render op not supported\n",
+ __FUNCTION__));
+ return false;
+ }
+
+ dx *= SAMPLES_X;
+ dy *= SAMPLES_Y;
+ if (!tor_init(&tor, &extents, 2*ntrap))
+ goto skip;
+
+ for (n = 0; n < ntrap; n++) {
+ xPointFixed p1, p2;
+
+ if (pixman_fixed_to_int(trap[n].top.y) + dst->pDrawable->y >= extents.y2 ||
+ pixman_fixed_to_int(trap[n].bot.y) + dst->pDrawable->y < extents.y1)
+ continue;
+
+ p1.y = trap[n].top.y;
+ p2.y = trap[n].bot.y;
+ p1.x = trap[n].top.l;
+ p2.x = trap[n].bot.l;
+ polygon_add_line(tor.polygon, &p1, &p2, dx, dy);
+
+ p1.y = trap[n].bot.y;
+ p2.y = trap[n].top.y;
+ p1.x = trap[n].top.r;
+ p2.x = trap[n].bot.r;
+ polygon_add_line(tor.polygon, &p1, &p2, dx, dy);
+ }
+
+ tor_render(sna, &tor, &tmp, clip,
+ choose_span(&tmp, dst, NULL, clip), false);
+
+ tor_fini(&tor);
+skip:
+ tmp.done(sna, &tmp);
+ return true;
+}
commit b4872f74c05a92c91d7b57146eb35ed2ecc86d97
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Tue Oct 21 10:38:22 2014 +0100
sna: Avoid pwriting large batches
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/kgem.c b/src/sna/kgem.c
index 234d1d0..25ed0ad 100644
--- a/src/sna/kgem.c
+++ b/src/sna/kgem.c
@@ -1252,6 +1252,99 @@ out:
gem_close(kgem->fd, tiling.handle);
}
+static void kgem_fixup_relocs(struct kgem *kgem, struct kgem_bo *bo, int shrink)
+{
+ int n;
+
+ DBG(("%s: shrinking by %d\n", __FUNCTION__, shrink));
+
+ bo->target_handle = kgem->has_handle_lut ? kgem->nexec : bo->handle;
+
+ assert(kgem->nreloc__self <= 256);
+ if (kgem->nreloc__self) {
+ for (n = 0; n < kgem->nreloc__self; n++) {
+ int i = kgem->reloc__self[n];
+ assert(kgem->reloc[i].target_handle == ~0U);
+ kgem->reloc[i].target_handle = bo->target_handle;
+ kgem->reloc[i].presumed_offset = bo->presumed_offset;
+ kgem->batch[kgem->reloc[i].offset/sizeof(kgem->batch[0])] =
+ kgem->reloc[i].delta + bo->presumed_offset - shrink;
+ }
+
+ if (n == 256) {
+ for (n = kgem->reloc__self[255]; n < kgem->nreloc; n++) {
+ if (kgem->reloc[n].target_handle == ~0U) {
+ kgem->reloc[n].target_handle = bo->target_handle;
+ kgem->reloc[n].presumed_offset = bo->presumed_offset;
+ kgem->batch[kgem->reloc[n].offset/sizeof(kgem->batch[0])] =
+ kgem->reloc[n].delta + bo->presumed_offset - shrink;
+ }
+ }
+ }
+ }
+
+ if (shrink) {
+ for (n = 0; n < kgem->nreloc; n++) {
+ if (kgem->reloc[n].read_domains == I915_GEM_DOMAIN_INSTRUCTION &&
+ kgem->reloc[n].target_handle == bo->target_handle)
+ kgem->reloc[n].delta -= shrink;
+
+ if (kgem->reloc[n].offset >= sizeof(uint32_t)*kgem->nbatch)
+ kgem->reloc[n].offset -= shrink;
+ }
+ }
+}
+
+static struct kgem_bo *kgem_new_batch(struct kgem *kgem)
+{
+ struct kgem_bo *last;
+ unsigned flags;
+
+ last = kgem->batch_bo;
+ if (last) {
+ kgem_fixup_relocs(kgem, last, 0);
+ kgem->batch = NULL;
+ }
+
+ if (kgem->batch) {
+ assert(last == NULL);
+ return NULL;
+ }
+
+ flags = CREATE_CPU_MAP | CREATE_NO_THROTTLE;
+ if (!kgem->has_llc)
+ flags |= CREATE_UNCACHED;
+
+ kgem->batch_bo = kgem_create_linear(kgem,
+ sizeof(uint32_t)*kgem->batch_size,
+ flags);
+ if (kgem->batch_bo)
+ kgem->batch = kgem_bo_map__cpu(kgem, kgem->batch_bo);
+ if (kgem->batch == NULL) {
+ DBG(("%s: unable to map batch bo, mallocing(size=%d)\n",
+ __FUNCTION__,
+ sizeof(uint32_t)*kgem->batch_size));
+ if (kgem->batch_bo) {
+ kgem_bo_destroy(kgem, kgem->batch_bo);
+ kgem->batch_bo = NULL;
+ }
+
+ if (posix_memalign((void **)&kgem->batch, PAGE_SIZE,
+ ALIGN(sizeof(uint32_t) * kgem->batch_size, PAGE_SIZE))) {
+ ERR(("%s: batch allocation failed, disabling acceleration\n", __FUNCTION__));
+ __kgem_set_wedged(kgem);
+ }
+ } else {
+ DBG(("%s: allocated and mapped batch handle=%d [size=%d]\n",
+ __FUNCTION__, kgem->batch_bo->handle,
+ sizeof(uint32_t)*kgem->batch_size));
+ kgem_bo_sync__cpu(kgem, kgem->batch_bo);
+ }
+
+ DBG(("%s: using last batch handle=%d\n",
+ __FUNCTION__, last ? last->handle : 0));
+ return last;
+}
void kgem_init(struct kgem *kgem, int fd, struct pci_device *dev, unsigned gen)
{
@@ -1362,7 +1455,7 @@ void kgem_init(struct kgem *kgem, int fd, struct pci_device *dev, unsigned gen)
kgem->wedged = 1;
}
- kgem->batch_size = ARRAY_SIZE(kgem->batch);
+ kgem->batch_size = UINT16_MAX & ~7;
if (gen == 020 && !kgem->has_pinned_batches)
/* Limited to what we can pin */
kgem->batch_size = 4*1024;
@@ -1382,6 +1475,7 @@ void kgem_init(struct kgem *kgem, int fd, struct pci_device *dev, unsigned gen)
DBG(("%s: maximum batch size? %d\n", __FUNCTION__,
kgem->batch_size));
+ kgem_new_batch(kgem);
kgem->half_cpu_cache_pages = cpu_cache_size() >> 13;
DBG(("%s: last-level cache size: %d bytes, threshold in pages: %d\n",
@@ -1803,36 +1897,6 @@ static uint32_t kgem_end_batch(struct kgem *kgem)
return kgem->nbatch;
}
-static void kgem_fixup_self_relocs(struct kgem *kgem, struct kgem_bo *bo)
-{
- int n;
-
- assert(kgem->nreloc__self <= 256);
- if (kgem->nreloc__self == 0)
- return;
-
- for (n = 0; n < kgem->nreloc__self; n++) {
- int i = kgem->reloc__self[n];
- assert(kgem->reloc[i].target_handle == ~0U);
- kgem->reloc[i].target_handle = bo->target_handle;
- kgem->reloc[i].presumed_offset = bo->presumed_offset;
- kgem->batch[kgem->reloc[i].offset/sizeof(kgem->batch[0])] =
- kgem->reloc[i].delta + bo->presumed_offset;
- }
-
- if (n == 256) {
- for (n = kgem->reloc__self[255]; n < kgem->nreloc; n++) {
- if (kgem->reloc[n].target_handle == ~0U) {
- kgem->reloc[n].target_handle = bo->target_handle;
- kgem->reloc[n].presumed_offset = bo->presumed_offset;
- kgem->batch[kgem->reloc[n].offset/sizeof(kgem->batch[0])] =
- kgem->reloc[n].delta + bo->presumed_offset;
- }
- }
-
- }
-}
-
static void kgem_bo_binding_free(struct kgem *kgem, struct kgem_bo *bo)
{
struct kgem_bo_binding *b;
@@ -3041,15 +3105,16 @@ void kgem_reset(struct kgem *kgem)
kgem->needs_reservation = false;
kgem->flush = 0;
kgem->batch_flags = kgem->batch_flags_base;
+ assert(kgem->batch);
kgem->next_request = __kgem_request_alloc(kgem);
kgem_sna_reset(kgem);
}
-static int compact_batch_surface(struct kgem *kgem)
+static int compact_batch_surface(struct kgem *kgem, int *shrink)
{
- int size, shrink, n;
+ int size, n;
if (!kgem->has_relaxed_delta)
return kgem->batch_size * sizeof(uint32_t);
@@ -3059,29 +3124,22 @@ static int compact_batch_surface(struct kgem *kgem)
size = n - kgem->surface + kgem->nbatch;
size = ALIGN(size, 1024);
- shrink = n - size;
- if (shrink) {
- DBG(("shrinking from %d to %d\n", kgem->batch_size, size));
-
- shrink *= sizeof(uint32_t);
- for (n = 0; n < kgem->nreloc; n++) {
- if (kgem->reloc[n].read_domains == I915_GEM_DOMAIN_INSTRUCTION &&
- kgem->reloc[n].target_handle == ~0U)
- kgem->reloc[n].delta -= shrink;
-
- if (kgem->reloc[n].offset >= sizeof(uint32_t)*kgem->nbatch)
- kgem->reloc[n].offset -= shrink;
- }
- }
-
+ *shrink = (n - size) * sizeof(uint32_t);
return size * sizeof(uint32_t);
}
static struct kgem_bo *
-kgem_create_batch(struct kgem *kgem, int size)
+kgem_create_batch(struct kgem *kgem)
{
struct drm_i915_gem_set_domain set_domain;
struct kgem_bo *bo;
+ int shrink = 0;
+ int size;
+
+ if (kgem->surface != kgem->batch_size)
+ size = compact_batch_surface(kgem, &shrink);
+ else
+ size = kgem->nbatch * sizeof(uint32_t);
if (size <= 4096) {
bo = list_first_entry(&kgem->pinned_batches[0],
@@ -3091,7 +3149,8 @@ kgem_create_batch(struct kgem *kgem, int size)
out_4096:
assert(bo->refcnt > 0);
list_move_tail(&bo->list, &kgem->pinned_batches[0]);
- return kgem_bo_reference(bo);
+ bo = kgem_bo_reference(bo);
+ goto write;
}
if (!__kgem_busy(kgem, bo->handle)) {
@@ -3109,7 +3168,8 @@ out_4096:
out_16384:
assert(bo->refcnt > 0);
list_move_tail(&bo->list, &kgem->pinned_batches[1]);
- return kgem_bo_reference(bo);
+ bo = kgem_bo_reference(bo);
+ goto write;
}
if (!__kgem_busy(kgem, bo->handle)) {
@@ -3121,14 +3181,14 @@ out_16384:
if (kgem->gen == 020) {
bo = kgem_create_linear(kgem, size, CREATE_CACHED | CREATE_TEMPORARY);
if (bo)
- return bo;
+ goto write;
/* Nothing available for reuse, rely on the kernel wa */
if (kgem->has_pinned_batches) {
bo = kgem_create_linear(kgem, size, CREATE_CACHED | CREATE_TEMPORARY);
if (bo) {
kgem->batch_flags &= ~LOCAL_I915_EXEC_IS_PINNED;
- return bo;
+ goto write;
}
}
@@ -3152,11 +3212,26 @@ out_16384:
kgem_retire(kgem);
assert(bo->rq == NULL);
- return kgem_bo_reference(bo);
+ bo = kgem_bo_reference(bo);
+ goto write;
}
}
- return kgem_create_linear(kgem, size, CREATE_NO_THROTTLE);
+ bo = NULL;
+ if (!kgem->has_llc) {
+ bo = kgem_create_linear(kgem, size, CREATE_NO_THROTTLE);
+ if (bo) {
+write:
+ kgem_fixup_relocs(kgem, bo, shrink);
+ if (kgem_batch_write(kgem, bo->handle, size)) {
+ kgem_bo_destroy(kgem, bo);
+ return NULL;
+ }
+ }
+ }
+ if (bo == NULL)
+ bo = kgem_new_batch(kgem);
+ return bo;
}
#if !NDEBUG
@@ -3253,7 +3328,6 @@ void _kgem_submit(struct kgem *kgem)
{
struct kgem_request *rq;
uint32_t batch_end;
- int size;
assert(!DBG_NO_HW);
assert(!kgem->wedged);
@@ -3282,17 +3356,14 @@ void _kgem_submit(struct kgem *kgem)
__kgem_batch_debug(kgem, batch_end);
#endif
- if (kgem->surface != kgem->batch_size)
- size = compact_batch_surface(kgem);
- else
- size = kgem->nbatch * sizeof(kgem->batch[0]);
-
rq = kgem->next_request;
assert(rq->bo == NULL);
- rq->bo = kgem_create_batch(kgem, size);
+
+ rq->bo = kgem_create_batch(kgem);
if (rq->bo) {
+ struct drm_i915_gem_execbuffer2 execbuf;
uint32_t handle = rq->bo->handle;
- int i;
+ int i, ret;
assert(!rq->bo->needs_flush);
@@ -3306,113 +3377,105 @@ void _kgem_submit(struct kgem *kgem)
kgem->exec[i].rsvd1 = 0;
kgem->exec[i].rsvd2 = 0;
- rq->bo->target_handle = kgem->has_handle_lut ? i : handle;
rq->bo->exec = &kgem->exec[i];
rq->bo->rq = MAKE_REQUEST(rq, kgem->ring); /* useful sanity check */
list_add(&rq->bo->request, &rq->buffers);
rq->ring = kgem->ring == KGEM_BLT;
- kgem_fixup_self_relocs(kgem, rq->bo);
-
- if (kgem_batch_write(kgem, handle, size) == 0) {
- struct drm_i915_gem_execbuffer2 execbuf;
- int ret;
-
- memset(&execbuf, 0, sizeof(execbuf));
- execbuf.buffers_ptr = (uintptr_t)kgem->exec;
- execbuf.buffer_count = kgem->nexec;
- execbuf.batch_len = batch_end*sizeof(uint32_t);
- execbuf.flags = kgem->ring | kgem->batch_flags;
-
- if (DBG_DUMP) {
- int fd = open("/tmp/i915-batchbuffers.dump",
- O_WRONLY | O_CREAT | O_APPEND,
- 0666);
- if (fd != -1) {
- ret = write(fd, kgem->batch, batch_end*sizeof(uint32_t));
- fd = close(fd);
- }
+ memset(&execbuf, 0, sizeof(execbuf));
+ execbuf.buffers_ptr = (uintptr_t)kgem->exec;
+ execbuf.buffer_count = kgem->nexec;
+ execbuf.batch_len = batch_end*sizeof(uint32_t);
+ execbuf.flags = kgem->ring | kgem->batch_flags;
+
+ if (DBG_DUMP) {
+ int fd = open("/tmp/i915-batchbuffers.dump",
+ O_WRONLY | O_CREAT | O_APPEND,
+ 0666);
+ if (fd != -1) {
+ ret = write(fd, kgem->batch, batch_end*sizeof(uint32_t));
+ fd = close(fd);
}
+ }
- ret = do_execbuf(kgem, &execbuf);
- if (DEBUG_SYNC && ret == 0) {
- struct drm_i915_gem_set_domain set_domain;
+ ret = do_execbuf(kgem, &execbuf);
+ if (DEBUG_SYNC && ret == 0) {
+ struct drm_i915_gem_set_domain set_domain;
- VG_CLEAR(set_domain);
- set_domain.handle = handle;
- set_domain.read_domains = I915_GEM_DOMAIN_GTT;
- set_domain.write_domain = I915_GEM_DOMAIN_GTT;
+ VG_CLEAR(set_domain);
+ set_domain.handle = handle;
+ set_domain.read_domains = I915_GEM_DOMAIN_GTT;
+ set_domain.write_domain = I915_GEM_DOMAIN_GTT;
- ret = do_ioctl(kgem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &set_domain);
+ ret = do_ioctl(kgem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &set_domain);
+ }
+ if (ret < 0) {
+ kgem_throttle(kgem);
+ if (!kgem->wedged) {
+ xf86DrvMsg(kgem_get_screen_index(kgem), X_ERROR,
+ "Failed to submit rendering commands, disabling acceleration.\n");
+ __kgem_set_wedged(kgem);
}
- if (ret < 0) {
- kgem_throttle(kgem);
- if (!kgem->wedged) {
- xf86DrvMsg(kgem_get_screen_index(kgem), X_ERROR,
- "Failed to submit rendering commands, disabling acceleration.\n");
- __kgem_set_wedged(kgem);
- }
#if !NDEBUG
- ErrorF("batch[%d/%d]: %d %d %d, nreloc=%d, nexec=%d, nfence=%d, aperture=%d, fenced=%d, high=%d,%d: errno=%d\n",
- kgem->mode, kgem->ring, batch_end, kgem->nbatch, kgem->surface,
- kgem->nreloc, kgem->nexec, kgem->nfence, kgem->aperture, kgem->aperture_fenced, kgem->aperture_high, kgem->aperture_total, -ret);
+ ErrorF("batch[%d/%d]: %d %d %d, nreloc=%d, nexec=%d, nfence=%d, aperture=%d, fenced=%d, high=%d,%d: errno=%d\n",
+ kgem->mode, kgem->ring, batch_end, kgem->nbatch, kgem->surface,
+ kgem->nreloc, kgem->nexec, kgem->nfence, kgem->aperture, kgem->aperture_fenced, kgem->aperture_high, kgem->aperture_total, -ret);
- for (i = 0; i < kgem->nexec; i++) {
- struct kgem_bo *bo, *found = NULL;
+ for (i = 0; i < kgem->nexec; i++) {
+ struct kgem_bo *bo, *found = NULL;
- list_for_each_entry(bo, &kgem->next_request->buffers, request) {
- if (bo->handle == kgem->exec[i].handle) {
- found = bo;
- break;
- }
+ list_for_each_entry(bo, &kgem->next_request->buffers, request) {
+ if (bo->handle == kgem->exec[i].handle) {
+ found = bo;
+ break;
}
- ErrorF("exec[%d] = handle:%d, presumed offset: %x, size: %d, tiling %d, fenced %d, snooped %d, deleted %d\n",
- i,
- kgem->exec[i].handle,
- (int)kgem->exec[i].offset,
- found ? kgem_bo_size(found) : -1,
- found ? found->tiling : -1,
- (int)(kgem->exec[i].flags & EXEC_OBJECT_NEEDS_FENCE),
- found ? found->snoop : -1,
- found ? found->purged : -1);
- }
- for (i = 0; i < kgem->nreloc; i++) {
- ErrorF("reloc[%d] = pos:%d, target:%d, delta:%d, read:%x, write:%x, offset:%x\n",
- i,
- (int)kgem->reloc[i].offset,
- kgem->reloc[i].target_handle,
- kgem->reloc[i].delta,
- kgem->reloc[i].read_domains,
- kgem->reloc[i].write_domain,
- (int)kgem->reloc[i].presumed_offset);
}
+ ErrorF("exec[%d] = handle:%d, presumed offset: %x, size: %d, tiling %d, fenced %d, snooped %d, deleted %d\n",
+ i,
+ kgem->exec[i].handle,
+ (int)kgem->exec[i].offset,
+ found ? kgem_bo_size(found) : -1,
+ found ? found->tiling : -1,
+ (int)(kgem->exec[i].flags & EXEC_OBJECT_NEEDS_FENCE),
+ found ? found->snoop : -1,
+ found ? found->purged : -1);
+ }
+ for (i = 0; i < kgem->nreloc; i++) {
+ ErrorF("reloc[%d] = pos:%d, target:%d, delta:%d, read:%x, write:%x, offset:%x\n",
+ i,
+ (int)kgem->reloc[i].offset,
+ kgem->reloc[i].target_handle,
+ kgem->reloc[i].delta,
+ kgem->reloc[i].read_domains,
+ kgem->reloc[i].write_domain,
+ (int)kgem->reloc[i].presumed_offset);
+ }
- {
- struct drm_i915_gem_get_aperture aperture;
- if (do_ioctl(kgem->fd, DRM_IOCTL_I915_GEM_GET_APERTURE, &aperture) == 0)
- ErrorF("Aperture size %lld, available %lld\n",
- (long long)aperture.aper_size,
- (long long)aperture.aper_available_size);
- }
+ {
+ struct drm_i915_gem_get_aperture aperture;
+ if (do_ioctl(kgem->fd, DRM_IOCTL_I915_GEM_GET_APERTURE, &aperture) == 0)
+ ErrorF("Aperture size %lld, available %lld\n",
+ (long long)aperture.aper_size,
+ (long long)aperture.aper_available_size);
+ }
- if (ret == -ENOSPC)
- dump_gtt_info(kgem);
- if (ret == -EDEADLK)
- dump_fence_regs(kgem);
-
- if (DEBUG_SYNC) {
- int fd = open("/tmp/batchbuffer", O_WRONLY | O_CREAT | O_APPEND, 0666);
- if (fd != -1) {
- int ignored = write(fd, kgem->batch, batch_end*sizeof(uint32_t));
- assert(ignored == batch_end*sizeof(uint32_t));
- close(fd);
- }
+ if (ret == -ENOSPC)
+ dump_gtt_info(kgem);
+ if (ret == -EDEADLK)
+ dump_fence_regs(kgem);
- FatalError("SNA: failed to submit batchbuffer, errno=%d\n", -ret);
+ if (DEBUG_SYNC) {
+ int fd = open("/tmp/batchbuffer", O_WRONLY | O_CREAT | O_APPEND, 0666);
+ if (fd != -1) {
+ int ignored = write(fd, kgem->batch, batch_end*sizeof(uint32_t));
+ assert(ignored == batch_end*sizeof(uint32_t));
+ close(fd);
}
-#endif
+
+ FatalError("SNA: failed to submit batchbuffer, errno=%d\n", -ret);
}
+#endif
}
}
#if SHOW_BATCH_AFTER
@@ -4167,16 +4230,18 @@ struct kgem_bo *kgem_create_linear(struct kgem *kgem, int size, unsigned flags)
}
size = NUM_PAGES(size);
- bo = search_linear_cache(kgem, size, CREATE_INACTIVE | flags);
- if (bo) {
- assert(bo->domain != DOMAIN_GPU);
- ASSERT_IDLE(kgem, bo->handle);
- bo->refcnt = 1;
- return bo;
- }
+ if ((flags & CREATE_UNCACHED) == 0) {
+ bo = search_linear_cache(kgem, size, CREATE_INACTIVE | flags);
+ if (bo) {
+ assert(bo->domain != DOMAIN_GPU);
+ ASSERT_IDLE(kgem, bo->handle);
+ bo->refcnt = 1;
+ return bo;
+ }
- if (flags & CREATE_CACHED)
- return NULL;
+ if (flags & CREATE_CACHED)
+ return NULL;
+ }
handle = gem_create(kgem->fd, size);
if (handle == 0)
diff --git a/src/sna/kgem.h b/src/sna/kgem.h
index 8bd5715..4a74f2e 100644
--- a/src/sna/kgem.h
+++ b/src/sna/kgem.h
@@ -120,12 +120,22 @@ enum {
};
struct kgem {
- int fd;
unsigned wedged;
+ int fd;
unsigned gen;
uint32_t unique_id;
+ uint16_t nbatch;
+ uint16_t surface;
+ uint16_t nexec;
+ uint16_t nreloc;
+ uint16_t nreloc__self;
+ uint16_t nfence;
+ uint16_t batch_size;
+
+ uint32_t *batch;
+
enum kgem_mode {
/* order matches I915_EXEC_RING ordering */
KGEM_NONE = 0,
@@ -158,14 +168,6 @@ struct kgem {
#define I915_EXEC_SECURE (1<<9)
#define LOCAL_EXEC_OBJECT_WRITE (1<<2)
- uint16_t nbatch;
- uint16_t surface;
- uint16_t nexec;
- uint16_t nreloc;
- uint16_t nreloc__self;
- uint16_t nfence;
- uint16_t batch_size;
-
uint32_t flush:1;
uint32_t need_expire:1;
uint32_t need_purge:1;
@@ -217,8 +219,9 @@ struct kgem {
int16_t dst_x, int16_t dst_y,
uint16_t width, uint16_t height);
+ struct kgem_bo *batch_bo;
+
uint16_t reloc__self[256];
- uint32_t batch[64*1024-8] page_aligned;
struct drm_i915_gem_exec_object2 exec[384] page_aligned;
struct drm_i915_gem_relocation_entry reloc[8192] page_aligned;
@@ -299,8 +302,9 @@ enum {
CREATE_PRIME = 0x20,
CREATE_TEMPORARY = 0x40,
CREATE_CACHED = 0x80,
- CREATE_NO_RETIRE = 0x100,
- CREATE_NO_THROTTLE = 0x200,
+ CREATE_UNCACHED = 0x100,
+ CREATE_NO_RETIRE = 0x200,
+ CREATE_NO_THROTTLE = 0x400,
};
struct kgem_bo *kgem_create_2d(struct kgem *kgem,
int width,
More information about the xorg-commit
mailing list