Fix vulkan image sampling

src/gpu/vk/GrVkGpu.cpp

Index: src/gpu/vk/GrVkGpu.cpp
diff --git a/src/gpu/vk/GrVkGpu.cpp b/src/gpu/vk/GrVkGpu.cpp
index c1cec9d96fa7ea77fc4d829566b56dd34bd6e0c0..e907d8d87860374ca2dd05a18b8ebccba01299a4 100644
--- a/src/gpu/vk/GrVkGpu.cpp
+++ b/src/gpu/vk/GrVkGpu.cpp
@@ -784,7 +784,8 @@ void GrVkGpu::generateMipmap(GrVkTexture* tex) const {
     // determine if we can blit to and from this format
     const GrVkCaps& caps = this->vkCaps();
     if (!caps.configCanBeDstofBlit(tex->config(), false) ||
-        !caps.configCanBeSrcofBlit(tex->config(), false)) {
+        !caps.configCanBeSrcofBlit(tex->config(), false) ||
+        !caps.mipMapSupport()) {
         return;
     }

[egdaniel, 2016/07/01 20:20:22]

On Nvidia, there seems to be a driver bug in blitting to create the mip maps. We
can land this CL as is which will occasionally draw corrupted triangles on on
parts of images we've mipmaped. This does not repro on the 5X. A work around
that seems to work is simply submitting the current command buffer here (doens't
need a sync) before blitting. We can wrap this is some cap that is set on Nvidia
only (but then we may not notice if their driver fixes it).

[bsalomon, 2016/07/02 00:50:12]

I leave it to you whether you want to workaround the bug or not. It'd be a good
to submit a bug to NV so they can fix it. Then we'd also probably know when to
remove the workaround after the fix.

[jvanverth1, 2016/07/06 15:19:42]

I'm fine with landing this as is, and creating a separate patch that submits the
command buffer.