Skip to content

S
swiftshader
Switch branch/tag
  1. 11 Dec, 2019 3 commits
  3. 10 Dec, 2019 15 commits
  5. 09 Dec, 2019 13 commits
  7. 07 Dec, 2019 1 commit
  9. 06 Dec, 2019 6 commits
  11. 05 Dec, 2019 2 commits
    • [vulkan] Add VK_SUBGROUP_FEATURE_ARITHMETIC_BIT support · b9f03f47
      David 'Digit' Turner authored 
      As the name suggest, this CL adds many operations related
to arithmetic subgroup operations, i.e.:

  - Reduction / Inclusive Scan / Exclusive Scans for:

    - IntAdd, UIntAdd, FloatAdd
    - IntMul, UIntMul, FloatMul,
    - IntMin, UIntMin, FloatMin,
    - IntMax, UIntMax, FloatMax,
    - BitwiseAnd, BitwiseOr, BitwiseXor
    - LogicalAnd, LogicalOr, LogicalXor

The implementation uses a single template function to
implement all these, based on the fact that these are
all binary commutative operations.

NOTE: Only 32-bit values are supported.

To make scans efficient, a new Reactor operation, named
Blend() is introduced. It is used to mix two input vectors
using 4 3-bit indices (encoded in a single 16-bit value)
to select the result's lane values.

A new unit-test is added to ReactorUnittests to check its
behaviour. Unfortunately, the test takes about 2 minutes
on a fast workstation when doing a full scan, so it will
by default only check 1/11th of all possible values
(see comments in the patch for more details).

Also, Float4::positive_inf() and Float4::negative_inf()
methods were added, since trying to build Float4(INFINITY)
will trigger a DCHECK() in the Float4(float) constructor,
and the infinity values are required by the subgroup
floating-point scan operations.

Bug: b/142002682
Test: dEQP-VK.subgroups.arithmetic.*
Change-Id: I86f509fc47f7475ca126615ed698ee493ae835ef
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38929Reviewed-by: 's avatarChris Forbes <chrisforbes@google.com>
Reviewed-by: 's avatarBen Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: 's avatarDavid Turner <digit@google.com>
    • Vulkan: Optimize shaderClipDistance and shaderCullDistance · f6a128b6
      Ben Clayton authored 
      Only process clip/cull distances if they're actually outputted from the vertex shader.

There's still overhead of having these distances stored in each vertex / primitive, but we should optimize the structure sizes as a larger set of changes.

Bug: b/139207336
Tests: dEQP-VK.clipping.*
Change-Id: I8f04b1c3ea823bb1a8cf62f18c987e01cd0c979a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35032Reviewed-by: 's avatarChris Forbes <chrisforbes@google.com>
Tested-by: 's avatarBen Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>