AuroraMiddleware/SPIRV-Cross
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
781367d083
SPIRV-Cross/reference/shaders-hlsl-no-opt/comp/subgroups.invalid.nofxc.sm60.comp

96 lines
4.6 KiB
Plaintext
Raw Normal View History

Update external/ to SPIR-V 1.5 Rolled the hashes used for glslang, SPIRV-Tools, and SPIRV-Headers to HEAD, which includes the update to 1.5. Added passing '--amb' to glslang, so I didn't have to explicitly set bindings in a large number of test shaders that currently don't, and now glslang considers them invalid. Marked all shaders that no longer pass spirv-val as .invalid.
2019-09-18 19:56:51 +00:00
static const uint3 gl_WorkGroupSize = uint3(1u, 1u, 1u);
Partially implement subgroup ops for HLSL SM 6.0. Lots of stuff that needs tons of emulation, which I'm not going to bother with.
2018-04-11 13:02:02 +00:00
RWByteAddressBuffer _9 : register(u0, space0);
static uint4 gl_SubgroupEqMask;
static uint4 gl_SubgroupGeMask;
static uint4 gl_SubgroupGtMask;
static uint4 gl_SubgroupLeMask;
static uint4 gl_SubgroupLtMask;
void comp_main()
{
_9.Store(0, asuint(float(WaveGetLaneCount())));
_9.Store(0, asuint(float(WaveGetLaneIndex())));
Only run spirv-opt if the spirv is valid. This CL updates the test runner to only run spirv-opt if the generated SPIR-V is valid. If validation is skipped it's possible to hit aborts and other memory errors in the optimizer as it assumes the SPIR-V is valid.
2019-11-05 15:59:41 +00:00
bool elected = WaveIsFirstLane();
Partially implement subgroup ops for HLSL SM 6.0. Lots of stuff that needs tons of emulation, which I'm not going to bother with.
2018-04-11 13:02:02 +00:00
_9.Store(0, asuint(float4(gl_SubgroupEqMask).x));
_9.Store(0, asuint(float4(gl_SubgroupGeMask).x));
_9.Store(0, asuint(float4(gl_SubgroupGtMask).x));
_9.Store(0, asuint(float4(gl_SubgroupLeMask).x));
_9.Store(0, asuint(float4(gl_SubgroupLtMask).x));
Only run spirv-opt if the spirv is valid. This CL updates the test runner to only run spirv-opt if the generated SPIR-V is valid. If validation is skipped it's possible to hit aborts and other memory errors in the optimizer as it assumes the SPIR-V is valid.
2019-11-05 15:59:41 +00:00
float4 broadcasted = WaveReadLaneAt(10.0f.xxxx, 8u);
float3 first = WaveReadLaneFirst(20.0f.xxx);
uint4 ballot_value = WaveActiveBallot(true);
uint bit_count = countbits(ballot_value.x) + countbits(ballot_value.y) + countbits(ballot_value.z) + countbits(ballot_value.w);
bool has_all = WaveActiveAllTrue(true);
bool has_any = WaveActiveAnyTrue(true);
bool has_equal = WaveActiveAllEqualBool(true);
float4 added = WaveActiveSum(20.0f.xxxx);
int4 iadded = WaveActiveSum(int4(20, 20, 20, 20));
float4 multiplied = WaveActiveProduct(20.0f.xxxx);
int4 imultiplied = WaveActiveProduct(int4(20, 20, 20, 20));
float4 lo = WaveActiveMin(20.0f.xxxx);
float4 hi = WaveActiveMax(20.0f.xxxx);
int4 slo = WaveActiveMin(int4(20, 20, 20, 20));
int4 shi = WaveActiveMax(int4(20, 20, 20, 20));
uint4 ulo = WaveActiveMin(uint4(20u, 20u, 20u, 20u));
uint4 uhi = WaveActiveMax(uint4(20u, 20u, 20u, 20u));
uint4 anded = WaveActiveBitAnd(ballot_value);
uint4 ored = WaveActiveBitOr(ballot_value);
uint4 xored = WaveActiveBitXor(ballot_value);
added = WavePrefixSum(added) + added;
iadded = WavePrefixSum(iadded) + iadded;
multiplied = WavePrefixProduct(multiplied) * multiplied;
imultiplied = WavePrefixProduct(imultiplied) * imultiplied;
added = WavePrefixSum(multiplied);
multiplied = WavePrefixProduct(multiplied);
iadded = WavePrefixSum(imultiplied);
imultiplied = WavePrefixProduct(imultiplied);
float4 swap_horiz = QuadReadAcrossX(20.0f.xxxx);
float4 swap_vertical = QuadReadAcrossY(20.0f.xxxx);
float4 swap_diagonal = QuadReadAcrossDiagonal(20.0f.xxxx);
float4 quad_broadcast = QuadReadLaneAt(20.0f.xxxx, 3u);
Partially implement subgroup ops for HLSL SM 6.0. Lots of stuff that needs tons of emulation, which I'm not going to bother with.
2018-04-11 13:02:02 +00:00
}
[numthreads(1, 1, 1)]
void main()
{
gl_SubgroupEqMask = 1u << (WaveGetLaneIndex() - uint4(0, 32, 64, 96));
if (WaveGetLaneIndex() >= 32) gl_SubgroupEqMask.x = 0;
if (WaveGetLaneIndex() >= 64 || WaveGetLaneIndex() < 32) gl_SubgroupEqMask.y = 0;
if (WaveGetLaneIndex() >= 96 || WaveGetLaneIndex() < 64) gl_SubgroupEqMask.z = 0;
if (WaveGetLaneIndex() < 96) gl_SubgroupEqMask.w = 0;
gl_SubgroupGeMask = ~((1u << (WaveGetLaneIndex() - uint4(0, 32, 64, 96))) - 1u);
if (WaveGetLaneIndex() >= 32) gl_SubgroupGeMask.x = 0u;
if (WaveGetLaneIndex() >= 64) gl_SubgroupGeMask.y = 0u;
if (WaveGetLaneIndex() >= 96) gl_SubgroupGeMask.z = 0u;
if (WaveGetLaneIndex() < 32) gl_SubgroupGeMask.y = ~0u;
if (WaveGetLaneIndex() < 64) gl_SubgroupGeMask.z = ~0u;
if (WaveGetLaneIndex() < 96) gl_SubgroupGeMask.w = ~0u;
uint gt_lane_index = WaveGetLaneIndex() + 1;
gl_SubgroupGtMask = ~((1u << (gt_lane_index - uint4(0, 32, 64, 96))) - 1u);
if (gt_lane_index >= 32) gl_SubgroupGtMask.x = 0u;
if (gt_lane_index >= 64) gl_SubgroupGtMask.y = 0u;
if (gt_lane_index >= 96) gl_SubgroupGtMask.z = 0u;
if (gt_lane_index >= 128) gl_SubgroupGtMask.w = 0u;
if (gt_lane_index < 32) gl_SubgroupGtMask.y = ~0u;
if (gt_lane_index < 64) gl_SubgroupGtMask.z = ~0u;
if (gt_lane_index < 96) gl_SubgroupGtMask.w = ~0u;
uint le_lane_index = WaveGetLaneIndex() + 1;
gl_SubgroupLeMask = (1u << (le_lane_index - uint4(0, 32, 64, 96))) - 1u;
if (le_lane_index >= 32) gl_SubgroupLeMask.x = ~0u;
if (le_lane_index >= 64) gl_SubgroupLeMask.y = ~0u;
if (le_lane_index >= 96) gl_SubgroupLeMask.z = ~0u;
if (le_lane_index >= 128) gl_SubgroupLeMask.w = ~0u;
if (le_lane_index < 32) gl_SubgroupLeMask.y = 0u;
if (le_lane_index < 64) gl_SubgroupLeMask.z = 0u;
if (le_lane_index < 96) gl_SubgroupLeMask.w = 0u;
gl_SubgroupLtMask = (1u << (WaveGetLaneIndex() - uint4(0, 32, 64, 96))) - 1u;
if (WaveGetLaneIndex() >= 32) gl_SubgroupLtMask.x = ~0u;
if (WaveGetLaneIndex() >= 64) gl_SubgroupLtMask.y = ~0u;
if (WaveGetLaneIndex() >= 96) gl_SubgroupLtMask.z = ~0u;
if (WaveGetLaneIndex() < 32) gl_SubgroupLtMask.y = 0u;
if (WaveGetLaneIndex() < 64) gl_SubgroupLtMask.z = 0u;
if (WaveGetLaneIndex() < 96) gl_SubgroupLtMask.w = 0u;
comp_main();
}
Reference in New Issue Copy Permalink