tree 41a1f79a5b96a2ec469bc271b8e159bdcf703800
parent c471ec24de521f338db464794f1edd9185ad38bf
author sbarati@apple.com <sbarati@apple.com@268f45cc-cd09-0410-ab3c-d52691b4dbfc> 1563249025 +0000
committer sbarati@apple.com <sbarati@apple.com@268f45cc-cd09-0410-ab3c-d52691b4dbfc> 1563249025 +0000

[WHLSL] Matrix memory layout should match HLSL by laying out columns linearly
https://bugs.webkit.org/show_bug.cgi?id=199215

Reviewed by Myles C. Maxfield.

Source/WebCore:

This patch makes it so that we lay out matrices in memory in the same
way HLSL does. This is by laying out columns linearly in memory. So a float4x4
composed by this series of floats in memory:
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]

composes this logical matrix:
[[ 0,  4,  8, 12]
 [ 1,  5,  9, 13]
 [ 2,  6, 10, 14]
 [ 3,  7, 11, 15]]

To implement this, we switch to using an array to represent the memory
contents linear memory layout of a matrix. So the matrix float4x3 will now
be an array<float, 12> in metal. Then, we change the indexed getter and
setter methods for matrices to load and store from and to the correct
memory locations. The memory layout of matrices is observable to WHLSL
when using a matrix as an input/output to a shader.

Test: webgpu/whlsl-matrix-memory-layout.html

* Modules/webgpu/WHLSL/Metal/WHLSLNativeFunctionWriter.cpp:
(WebCore::WHLSL::Metal::writeNativeFunction):
* Modules/webgpu/WHLSL/Metal/WHLSLNativeTypeWriter.cpp:
(WebCore::WHLSL::Metal::writeNativeType):

LayoutTests:

* webgpu/whlsl-matrix-memory-layout-expected.txt: Added.
* webgpu/whlsl-matrix-memory-layout.html: Added.
* webgpu/whlsl-test-harness-test.html:


git-svn-id: http://svn.webkit.org/repository/webkit/trunk@247468 268f45cc-cd09-0410-ab3c-d52691b4dbfc