v1 - had problems with msan and unintialized glyphs.
v2 - had problems with typefaces with no glyphs in them
This adds a check to make sure there are glyphs in the font
when going to uniquify.
Change-Id: Id27fa4578be33da1e468b4652db19740ddcadfc6
Reviewed-on: https://skia-review.googlesource.com/140785
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Change-Id: I0204a9522e828c87bb7c6c20ae34ce51161442af
Reviewed-on: https://skia-review.googlesource.com/137895
Reviewed-by: Herb Derby <herb@google.com>
Reviewed-by: Ben Wagner <bungeman@google.com>
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Ben Wagner <bungeman@google.com>
This is a reland of f4c5bb9aba
Original change's description:
> Make GPU cache invalidation SkMessageBus messages go to one GrContext.
>
> Makes it so the template param to SkMessageBus must implement:
> bool shouldSend(uint32_t inboxID) const
>
> Updates all GPU backend message types to only go to the GrContext that
> is adding a cache entry.
>
> Bug: skia:
> Change-Id: I3e8a4eb90654b7b8ac57cac9fb508c0ef1d51058
> Reviewed-on: https://skia-review.googlesource.com/140220
> Reviewed-by: Robert Phillips <robertphillips@google.com>
> Reviewed-by: Jim Van Verth <jvanverth@google.com>
Bug: skia:
Change-Id: I8402bfe3ed0170c99936d47050458817030b473b
Reviewed-on: https://skia-review.googlesource.com/140801
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
It turns out that SkDeviceProfile is no longer used and can just be
deleted. The ResourceCacheTest and DebugGLTestContext are changed to use
smart pointers where possible. This also clarifies the squirrelly part of
the test. DebugGLTestContext is going away soon anyway.
Change-Id: I95ef24afa58aa4d356429b93d4dec0d72e3fd827
Reviewed-on: https://skia-review.googlesource.com/140577
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Ben Wagner <bungeman@google.com>
parametric_a and gamma_dst were unused outside of unit tests.
In all other cases, we always use parametric_{r,g,b} together
and always pass them the same argument. So we can collapse
them into a single stage like gamma and to/from_srgb.
Change-Id: I08cea896c7744f97b4f4bf9e029f5d643e45e177
Reviewed-on: https://skia-review.googlesource.com/140576
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Brian Osman <brianosman@google.com>
This reverts commit 0421083a44.
Reason for revert: fushia page fault
Original change's description:
> Use new SkGlyphIDSet - v2
>
> Add bzero to make msan and valgrind happy.
>
> Change-Id: I9b4e2f2b8e690da4b4b920fef27d5a8854092219
> Reviewed-on: https://skia-review.googlesource.com/140563
> Reviewed-by: Mike Klein <mtklein@google.com>
> Commit-Queue: Herb Derby <herb@google.com>
TBR=mtklein@google.com,herb@google.com
Change-Id: I4e5e16644cbf56b5ff0b21afd6f3962e3976a1da
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/140803
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Herb Derby <herb@google.com>
This reverts commit b188da121c.
Reason for revert: Patch on a rollback
Original change's description:
> Use simple buffers instead of vectors
>
> Start using simple buffers, these will be used for
> multiple runs latter on.
>
> Change-Id: I8dadbed036b7a60d708c49b84bb5e3bb3710f704
> Reviewed-on: https://skia-review.googlesource.com/140578
> Reviewed-by: Mike Klein <mtklein@google.com>
> Commit-Queue: Herb Derby <herb@google.com>
TBR=mtklein@google.com,herb@google.com
Change-Id: I51d9ad9bc0fd7efb93e5db4dc504d8dc31e7cfb4
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/140802
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Herb Derby <herb@google.com>
This reverts commit f4c5bb9aba.
Reason for revert: Accidentally submitted without CQ
Original change's description:
> Make GPU cache invalidation SkMessageBus messages go to one GrContext.
>
> Makes it so the template param to SkMessageBus must implement:
> bool shouldSend(uint32_t inboxID) const
>
> Updates all GPU backend message types to only go to the GrContext that
> is adding a cache entry.
>
> Bug: skia:
> Change-Id: I3e8a4eb90654b7b8ac57cac9fb508c0ef1d51058
> Reviewed-on: https://skia-review.googlesource.com/140220
> Reviewed-by: Robert Phillips <robertphillips@google.com>
> Reviewed-by: Jim Van Verth <jvanverth@google.com>
TBR=jvanverth@google.com,bsalomon@google.com,robertphillips@google.com
Change-Id: Iadb667d8027341703d254325320ddaa528fb33a1
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:
Reviewed-on: https://skia-review.googlesource.com/140800
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Makes it so the template param to SkMessageBus must implement:
bool shouldSend(uint32_t inboxID) const
Updates all GPU backend message types to only go to the GrContext that
is adding a cache entry.
Bug: skia:
Change-Id: I3e8a4eb90654b7b8ac57cac9fb508c0ef1d51058
Reviewed-on: https://skia-review.googlesource.com/140220
Reviewed-by: Robert Phillips <robertphillips@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Start using simple buffers, these will be used for
multiple runs latter on.
Change-Id: I8dadbed036b7a60d708c49b84bb5e3bb3710f704
Reviewed-on: https://skia-review.googlesource.com/140578
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Change-Id: I6c4c4b43dfa6b59832c63f8fcf43192b4973d88b
Reviewed-on: https://skia-review.googlesource.com/140565
Commit-Queue: Mike Klein <mtklein@google.com>
Auto-Submit: Hal Canary <halcanary@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Change-Id: I9ca830b2bcf3341b06bc4e261dba40809b452194
Reviewed-on: https://skia-review.googlesource.com/140572
Commit-Queue: Brian Osman <brianosman@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Auto-Submit: Brian Osman <brianosman@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
* Add fuzzer
* Add bench tests
* Add additional unit test
* Fix some bugs these exposed.
Bug: skia:
Change-Id: I6c587c92cb6cff32ab8300020b78f9f247d2bf64
Reviewed-on: https://skia-review.googlesource.com/139169
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Add bzero to make msan and valgrind happy.
Change-Id: I9b4e2f2b8e690da4b4b920fef27d5a8854092219
Reviewed-on: https://skia-review.googlesource.com/140563
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Herb Derby <herb@google.com>
If an edge is unsortable, its winding value is untrustworthy;
it cannot be added to the output on that alone. If one end
adjoins a trusted edge, and that edge is added to the output,
the untrusted edge can be added as well.
Also add in msvs debugging for angles.
With this change, all extended tests pass.
TBR=reed@google.com
Bug: skia:8125
Change-Id: I049c6efa2fa83edd7b49cdd598ec94c356481b0f
Reviewed-on: https://skia-review.googlesource.com/140562
Commit-Queue: Cary Clark <caryclark@skia.org>
Auto-Submit: Cary Clark <caryclark@skia.org>
Reviewed-by: Cary Clark <caryclark@skia.org>
The most common use-case for F16 is to draw things that are explicitly
outside of [0,1]. We can't prevent out-of-range values in general, users
could just draw out-of-range content in src mode. We've decided that
it's best to avoid surprising behavior, and trust the user in these
situations.
This effectively reverts https://skia-review.googlesource.com/c/skia/+/48183
Bug: skia:
Change-Id: I71c2dd925c567e2862d96ab70c752f19245a69c4
Reviewed-on: https://skia-review.googlesource.com/140500
Reviewed-by: Mike Klein <mtklein@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
This led to removing a lot of transfer function behavior code. There is
more that could be done, and we need to add in decoding to dst color
space, but this CL is almost entirely mechanical.
Change-Id: I91b2169f95aadcfaacdd2b9821bb1a01ce53f9a6
Reviewed-on: https://skia-review.googlesource.com/140349
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
This reverts commit 819f73c23c.
Reason for revert: uninitialized memory - this is expected but
Original change's description:
> Use new SkGlyphIDSet
>
> Change-Id: I6b8080393a22a56577528f66630ad39372edf712
> Reviewed-on: https://skia-review.googlesource.com/140243
> Commit-Queue: Herb Derby <herb@google.com>
> Reviewed-by: Mike Klein <mtklein@google.com>
TBR=mtklein@google.com,herb@google.com
Change-Id: I43e204520710738e9e8c84b0eb00260ca06fe6a2
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/140384
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Change-Id: I6b8080393a22a56577528f66630ad39372edf712
Reviewed-on: https://skia-review.googlesource.com/140243
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Test filinmangust14 exposes two problems:
- finding top of contour can expose divide by zero
- joining partial contour results can add diagonal
The latter makes the test return the wrong result,
and has not been seen in other tests. The fix
is to join disconnected contours by only following
the contours provided as input. Working on that.
The former bug is more straight-forward; just
don't try to compute axis-aligned intersection
if the denominator is zero.
All existing tests prior to the new one work
with this change.
R=caryclark@google.com
Bug: skia:8125
Change-Id: Ic878d090066708d9baca8475f27d4d5aba2294cc
Reviewed-on: https://skia-review.googlesource.com/140121
Reviewed-by: Cary Clark <caryclark@skia.org>
Commit-Queue: Cary Clark <caryclark@skia.org>
Auto-Submit: Cary Clark <caryclark@skia.org>
Temporarily remove run list code to gut some
overly complicated code.
Change-Id: Ib12efc394c05dee391143b440b2fab5bba4f22ae
Reviewed-on: https://skia-review.googlesource.com/139865
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Removes circular dependencies from coming change.
Some IWYU fallout from removing #includes from GrPathRenderer.h and
GrDrawingManager.h
Bug: skia:
Change-Id: Ice377538e6d64b6a74a59e6140e1de9a58ab99bf
Reviewed-on: https://skia-review.googlesource.com/140181
Commit-Queue: Brian Salomon <bsalomon@google.com>
Auto-Submit: Brian Salomon <bsalomon@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
Bug: skia:
Change-Id: Ic0e0ae23c88f86885583a304d90d1ce874ff14ed
Reviewed-on: https://skia-review.googlesource.com/139960
Auto-Submit: Brian Osman <brianosman@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Don't try to guess the pixel config to use for intermediates. Instead,
just make the intermediates in the same color type (and space) as the
final destination. This removes some no-longer-correct logic that was
using sRGB configs, resulting in linear blending and precision loss.
Change-Id: I627c47193a9f2889c3dc121170ff3e7d5d315fa0
Reviewed-on: https://skia-review.googlesource.com/139547
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Also:
* clean up PolyUtils checks to be correct and consistent.
* fix some bugs discovered by the unit tests.
Bug: skia:
Change-Id: I1a8e07d13cb44fecc67344154dc1002f3f910f5d
Reviewed-on: https://skia-review.googlesource.com/138592
Reviewed-by: Robert Phillips <robertphillips@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
In some cases, splitting may produce three consecutive edges which
are collinear. The first one was being merged out, causing the third
one to be missed.
The fix is to switch the arguments to merge_edges_*, ensuring that the
second parameter (the destination edge) is never merged out.
Bug: 851409
Change-Id: I65be2e8222846c99f7bc8d17ea61ddead617cc31
Reviewed-on: https://skia-review.googlesource.com/138700
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Stephen White <senorblanco@chromium.org>
All of the restrictions/assumptions that led to this code are gone,
so we can always use appropriate color space.
For the YUV provider, if/when we re-introduce 8888 sRGB, the color
space will have a linear transfer function, so the color space
xform will automatically do what was happening here. That removes
the last usage of framebuffer sRGB control, so we can remove all
kinds of GrPaint and GrPipeline plumbing for that feature.
Change-Id: I24af1d498dbc75210f92f8c61b10aa31eec022f6
Reviewed-on: https://skia-review.googlesource.com/138986
Reviewed-by: Mike Klein <mtklein@chromium.org>
Commit-Queue: Brian Osman <brianosman@google.com>
This reverts commit a219419c9d.
Reason for revert: breaking things
Original change's description:
> Some scissor state cleanup.
>
> Separate flushing the enablement of scissor from the rect in GrGLGpu.
>
> Move GrPipeline::ScissorState to a global enum and use more broadly.
> Rename to GrScissorTest to avoid name conflict with existing
> GrScissorState.
>
> Change-Id: Ib32160b3300bc12de2d2e1761d152fd1bba8b683
> Reviewed-on: https://skia-review.googlesource.com/137395
> Commit-Queue: Brian Salomon <bsalomon@google.com>
> Reviewed-by: Chris Dalton <csmartdalton@google.com>
TBR=bsalomon@google.com,csmartdalton@google.com
Change-Id: If71a5c5efc86d4239b40675bad2a6cb1f77460f8
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/138900
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Separate flushing the enablement of scissor from the rect in GrGLGpu.
Move GrPipeline::ScissorState to a global enum and use more broadly.
Rename to GrScissorTest to avoid name conflict with existing
GrScissorState.
Change-Id: Ib32160b3300bc12de2d2e1761d152fd1bba8b683
Reviewed-on: https://skia-review.googlesource.com/137395
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Chris Dalton <csmartdalton@google.com>
This reverts commit 26bb0e66f2.
Reason for revert: segfaulty
Here's a log with a stacktrace:
https://chromium-swarm.appspot.com/task?id=3e658f7aa40dbb10&refresh=10
Original change's description:
> GrTessellator: handle three consecutive collinear edges.
>
> In some cases, splitting may produce three consecutive edges which
> are collinear. The first one was being merged out, causing the third
> one to be missed.
>
> The fix is to switch the arguments to merge_edges_*, ensuring that the
> second parameter (the destination edge) is never merged out.
>
> Bug: 851409.
> Change-Id: I70fbbc506e97a26b259c1443b6d1787adec0f9b0
> Reviewed-on: https://skia-review.googlesource.com/138561
> Reviewed-by: Robert Phillips <robertphillips@google.com>
> Commit-Queue: Stephen White <senorblanco@chromium.org>
TBR=robertphillips@google.com,senorblanco@chromium.org
Change-Id: I6ecfb4c487d6f96e9fae7b8b40d74162354ed57c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: 851409.
Reviewed-on: https://skia-review.googlesource.com/138640
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
In some cases, splitting may produce three consecutive edges which
are collinear. The first one was being merged out, causing the third
one to be missed.
The fix is to switch the arguments to merge_edges_*, ensuring that the
second parameter (the destination edge) is never merged out.
Bug: 851409.
Change-Id: I70fbbc506e97a26b259c1443b6d1787adec0f9b0
Reviewed-on: https://skia-review.googlesource.com/138561
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Stephen White <senorblanco@chromium.org>
Use this to fill concave shadows.
Bug: skia:7971
Change-Id: I63dc1ed845f9fa3fcd86f1ad13b03da23cae0313
Reviewed-on: https://skia-review.googlesource.com/135200
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
If the last vertex in a contour is collinear with its neighbours,
remove it (this edge case was missing). Otherwise, the simplify step
may try to split it indefinitely.
Bug: 851409
Change-Id: I7efa4e616cdc1508a73c7a9f3de9d3f571569af8
Reviewed-on: https://skia-review.googlesource.com/138106
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Stephen White <senorblanco@chromium.org>
This reverts commit e2e52e46ca.
Reason for revert: See if this is blocking the Chrome roll
Original change's description:
> Remove drawTextBlob from device use drawGlyphRunList
>
> Convert all backends to use GlyphRunList instead of
> text blobs. If the device did not originally implement
> drawTextBlob it will be simulated by drawPosText on the
> device.
>
> Other changes:
> Change to using an origin from absolulte positioning. The GPU
> code uses origin change to update blobs under translation.
>
> Change cluster to use const uint32_t instead of just
> uint32_t.
>
> Add SkPaint to runs.
>
> The draw filter is hosted up to the canavas level and applied there.
>
> Change-Id: Ib105b6bd26b67db55f1c954e37c79fbdcaa9d4a2
> Reviewed-on: https://skia-review.googlesource.com/137224
> Reviewed-by: Herb Derby <herb@google.com>
> Reviewed-by: Khusal Sagar <khushalsagar@chromium.org>
> Reviewed-by: Hal Canary <halcanary@google.com>
> Reviewed-by: Jim Van Verth <jvanverth@google.com>
> Commit-Queue: Herb Derby <herb@google.com>
TBR=jvanverth@google.com,halcanary@google.com,bungeman@google.com,herb@google.com,reed@google.com,khushalsagar@chromium.org,khushalsagar@google.com
Change-Id: I4d93a534990c89deee7d3aaa00ec40d47e0d2ece
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/138120
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
Convert all backends to use GlyphRunList instead of
text blobs. If the device did not originally implement
drawTextBlob it will be simulated by drawPosText on the
device.
Other changes:
Change to using an origin from absolulte positioning. The GPU
code uses origin change to update blobs under translation.
Change cluster to use const uint32_t instead of just
uint32_t.
Add SkPaint to runs.
The draw filter is hosted up to the canavas level and applied there.
Change-Id: Ib105b6bd26b67db55f1c954e37c79fbdcaa9d4a2
Reviewed-on: https://skia-review.googlesource.com/137224
Reviewed-by: Herb Derby <herb@google.com>
Reviewed-by: Khusal Sagar <khushalsagar@chromium.org>
Reviewed-by: Hal Canary <halcanary@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Herb Derby <herb@google.com>
* Clamp path polygon points to nearest 1/16th of a pixel to help
with floating point issues.
* Added check for multiple contour paths.
* Return empty SkVertices for certain degenerate cases to avoid
unnecessary blurs.
* Check iteration count in SkOffsetPolygon to avoid infinite loops.
* Add new tests to verify these.
Bug: skia:
Change-Id: Ie6ad48d2504e065dcc822609d369f90c56ef3ad3
Reviewed-on: https://skia-review.googlesource.com/136701
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Draws basically the same as f16.
The existing load_f32, load_f32_dst, and store_f32 stages all had the
same bug that we'd never noticed because dy was always 0 until now.
Change-Id: Ibbd393fa1acc5df414be4cdef0f5a9d11dcccdb3
Reviewed-on: https://skia-review.googlesource.com/137585
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Brian Osman <brianosman@google.com>
Reviewed-by: Mike Reed <reed@google.com>
Bug: skia:8076
Move SkImageGeneratorCG and SkImageGeneratorWIC to include/ports/,
and make them SK_API.
Remove SkImageGeneratorCG::NewFromEncodedCG. No known clients are using
it, and it was a private API in src/. There is already a
MakeFromEncodedCG to replace it.
Similarly, switch WIC from New to Make.
Add a compile test for using them with
SetImageGeneratorFromEncodedDataFactory.
Change-Id: I897eb7a887b3736a6c614a68e38f38b6a1942cf1
Reviewed-on: https://skia-review.googlesource.com/137387
Reviewed-by: Mike Reed <reed@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
Remove scissor rect from GrPipeline.
Draws can specify "fixed dynamic state" which doesn't use the dynamism at
all or can specify dynamic state arrays with an entry per GrMesh.
When we state other than scissor rects this will allow the caller to
use a mix of truly dynamic and fixed dynamic state. So a caller that
only has dynamic scissor rects doesn't need to store its remaining
unvarying state in an array.
Change-Id: I8fcc07eb600c72a26cc712b185755c2116021a8a
Reviewed-on: https://skia-review.googlesource.com/137223
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
jpeg_skip_scanlines was incorrectly incrementing an internal counter,
resulting in an infinite loop. (This only occurs for certain types of
progressive images, using certain sample sizes.)
The fix is at 26f109290d.
This is included in tip-of-tree, which is unofficially 2.0.0, so go ahead and
update to it.
Add a test based on the original bug.
Bug: b/78329453
Change-Id: I5ade9924812324d58668c26f71cd622ef93f40a7
Reviewed-on: https://skia-review.googlesource.com/129459
Commit-Queue: Leon Scroggins <scroggo@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
on mac.
Change-Id: I2b8777a5b54e190c37623ab798a89b4f1833ceac
Reviewed-on: https://skia-review.googlesource.com/137382
Commit-Queue: Bruce Wang <brucewang@google.com>
Reviewed-by: Ben Wagner <bungeman@google.com>
We were inconsistent about which order these were in. Having the processor first
will make the parameter order more logical for an upcoming change. Also, the primitive
processor comes logically before the pipeline.
Change-Id: I3968c5e4e6dff01f9c4ad311eb1795b3c7580ff5
Reviewed-on: https://skia-review.googlesource.com/137228
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
With multiple runs, the shared vectors were moving out from
under the pointers in earlier runs.
Change-Id: I486d2e603e18ea7effc0dbdbc7c5d3c545278703
Reviewed-on: https://skia-review.googlesource.com/137222
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Ben Wagner <bungeman@google.com>
