tests: Add test for waveop in infinite loop convergence.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This commit is contained in:
Hans-Kristian Arntzen
parent
55a6847c61
commit
ebe589d622
|
|
@ -4215,6 +4215,290 @@ void test_shader_sm64_packed(void)
|
|||
destroy_test_context(&context);
|
||||
}
|
||||
|
||||
void test_shader_waveop_maximal_convergence(void)
|
||||
{
|
||||
D3D12_ROOT_PARAMETER root_parameters[2];
|
||||
ID3D12PipelineState *pso_nonconverged;
|
||||
ID3D12PipelineState *pso_converged;
|
||||
D3D12_ROOT_SIGNATURE_DESC rs_desc;
|
||||
struct test_context context;
|
||||
struct resource_readback rb;
|
||||
ID3D12Resource *outputs;
|
||||
ID3D12Resource *inputs;
|
||||
unsigned int i;
|
||||
|
||||
static const uint32_t inputs_data[] =
|
||||
{
|
||||
2, 3, 1, 3,
|
||||
2, 0, 0, 1,
|
||||
0, 1, 3, 2,
|
||||
2, 1, 2, 2,
|
||||
};
|
||||
|
||||
static const uint32_t reference_converged[] =
|
||||
{
|
||||
25, 25, 25, 25,
|
||||
25, 25, 25, 25,
|
||||
25, 25, 25, 25,
|
||||
25, 25, 25, 25,
|
||||
};
|
||||
|
||||
static const uint32_t reference_nonconverged[] =
|
||||
{
|
||||
12, 9, 4, 9,
|
||||
12, 0, 0, 4,
|
||||
0, 4, 9, 12,
|
||||
12, 4, 12, 12,
|
||||
};
|
||||
|
||||
#if 0
|
||||
StructuredBuffer<uint> RO : register(t0);
|
||||
RWStructuredBuffer<uint> RW : register(u0);
|
||||
|
||||
[numthreads(16, 1, 1)]
|
||||
void main(uint thr : SV_DispatchThreadID)
|
||||
{
|
||||
uint v = RO[thr];
|
||||
uint result;
|
||||
while (true)
|
||||
{
|
||||
uint first = WaveReadLaneFirst(v);
|
||||
if (v == first)
|
||||
{
|
||||
result = WaveActiveSum(v);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
RW[thr] = result;
|
||||
}
|
||||
#endif
|
||||
|
||||
/* Compiled with Version: dxcompiler.dll: 1.5 - 1.4.1907.0; dxil.dll: 1.4(10.0.18362.1) */
|
||||
static const BYTE reconvergence_dxil[] =
|
||||
{
|
||||
0x44, 0x58, 0x42, 0x43, 0x27, 0x94, 0xf5, 0xbd, 0x53, 0x66, 0x70, 0xdb, 0xe2, 0x95, 0x5c, 0x8c, 0xdc, 0x10, 0x0b, 0xdf, 0x01, 0x00, 0x00, 0x00, 0xe4, 0x06, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00,
|
||||
0x34, 0x00, 0x00, 0x00, 0x44, 0x00, 0x00, 0x00, 0x54, 0x00, 0x00, 0x00, 0x64, 0x00, 0x00, 0x00, 0xc8, 0x00, 0x00, 0x00, 0x53, 0x46, 0x49, 0x30, 0x08, 0x00, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x49, 0x53, 0x47, 0x31, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x4f, 0x53, 0x47, 0x31, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x08, 0x00, 0x00, 0x00, 0x50, 0x53, 0x56, 0x30, 0x5c, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff, 0x05, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x44, 0x58, 0x49, 0x4c, 0x14, 0x06, 0x00, 0x00, 0x60, 0x00, 0x05, 0x00, 0x85, 0x01, 0x00, 0x00, 0x44, 0x58, 0x49, 0x4c, 0x00, 0x01, 0x00, 0x00,
|
||||
0x10, 0x00, 0x00, 0x00, 0xfc, 0x05, 0x00, 0x00, 0x42, 0x43, 0xc0, 0xde, 0x21, 0x0c, 0x00, 0x00, 0x7c, 0x01, 0x00, 0x00, 0x0b, 0x82, 0x20, 0x00, 0x02, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00,
|
||||
0x07, 0x81, 0x23, 0x91, 0x41, 0xc8, 0x04, 0x49, 0x06, 0x10, 0x32, 0x39, 0x92, 0x01, 0x84, 0x0c, 0x25, 0x05, 0x08, 0x19, 0x1e, 0x04, 0x8b, 0x62, 0x80, 0x14, 0x45, 0x02, 0x42, 0x92, 0x0b, 0x42,
|
||||
0xa4, 0x10, 0x32, 0x14, 0x38, 0x08, 0x18, 0x4b, 0x0a, 0x32, 0x52, 0x88, 0x48, 0x90, 0x14, 0x20, 0x43, 0x46, 0x88, 0xa5, 0x00, 0x19, 0x32, 0x42, 0xe4, 0x48, 0x0e, 0x90, 0x91, 0x22, 0xc4, 0x50,
|
||||
0x41, 0x51, 0x81, 0x8c, 0xe1, 0x83, 0xe5, 0x8a, 0x04, 0x29, 0x46, 0x06, 0x51, 0x18, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x1b, 0x8c, 0xe0, 0xff, 0xff, 0xff, 0xff, 0x07, 0x40, 0x02, 0xa8, 0x0d,
|
||||
0x84, 0xf0, 0xff, 0xff, 0xff, 0xff, 0x03, 0x20, 0x6d, 0x30, 0x86, 0xff, 0xff, 0xff, 0xff, 0x1f, 0x00, 0x09, 0xa8, 0x00, 0x49, 0x18, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x13, 0x82, 0x60, 0x42,
|
||||
0x20, 0x4c, 0x08, 0x06, 0x00, 0x00, 0x00, 0x00, 0x89, 0x20, 0x00, 0x00, 0x3d, 0x00, 0x00, 0x00, 0x32, 0x22, 0x48, 0x09, 0x20, 0x64, 0x85, 0x04, 0x93, 0x22, 0xa4, 0x84, 0x04, 0x93, 0x22, 0xe3,
|
||||
0x84, 0xa1, 0x90, 0x14, 0x12, 0x4c, 0x8a, 0x8c, 0x0b, 0x84, 0xa4, 0x4c, 0x10, 0x68, 0x23, 0x00, 0x25, 0x00, 0x14, 0xe6, 0x08, 0xc0, 0xa0, 0x0c, 0x63, 0x0c, 0x22, 0x73, 0x04, 0x48, 0x29, 0xc6,
|
||||
0x18, 0xc6, 0xd0, 0x21, 0x73, 0xcf, 0x70, 0xf9, 0x13, 0xf6, 0x10, 0x92, 0x1f, 0x02, 0xcd, 0xb0, 0x10, 0x28, 0x48, 0x73, 0x04, 0x41, 0x31, 0xd4, 0x30, 0x63, 0x2c, 0x62, 0x45, 0x01, 0x43, 0x8d,
|
||||
0x31, 0xc6, 0x18, 0x86, 0xdc, 0x4d, 0xc3, 0xe5, 0x4f, 0xd8, 0x43, 0x48, 0xfe, 0x4a, 0x48, 0x2b, 0x31, 0xf9, 0xc8, 0x6d, 0xa3, 0x62, 0x8c, 0x31, 0x46, 0x29, 0xe0, 0x50, 0x63, 0x50, 0x1c, 0x08,
|
||||
0x98, 0x89, 0x0c, 0xc6, 0x81, 0x1d, 0xc2, 0x61, 0x1e, 0xe6, 0xc1, 0x0d, 0x66, 0x81, 0x1e, 0xe4, 0xa1, 0x1e, 0xc6, 0x81, 0x1e, 0xea, 0x41, 0x1e, 0xca, 0x81, 0x1c, 0x44, 0xa1, 0x1e, 0xcc, 0xc1,
|
||||
0x1c, 0xca, 0x41, 0x1e, 0xf8, 0xa0, 0x1e, 0xdc, 0x61, 0x1e, 0xd2, 0xe1, 0x1c, 0xdc, 0xa1, 0x1c, 0xc8, 0x01, 0x0c, 0xd2, 0xc1, 0x1d, 0xe8, 0xc1, 0x0f, 0x50, 0x30, 0x88, 0xce, 0x64, 0x06, 0xe3,
|
||||
0xc0, 0x0e, 0xe1, 0x30, 0x0f, 0xf3, 0xe0, 0x06, 0xb2, 0x70, 0x0b, 0xb3, 0x40, 0x0f, 0xf2, 0x50, 0x0f, 0xe3, 0x40, 0x0f, 0xf5, 0x20, 0x0f, 0xe5, 0x40, 0x0e, 0xa2, 0x50, 0x0f, 0xe6, 0x60, 0x0e,
|
||||
0xe5, 0x20, 0x0f, 0x7c, 0x50, 0x0f, 0xee, 0x30, 0x0f, 0xe9, 0x70, 0x0e, 0xee, 0x50, 0x0e, 0xe4, 0x00, 0x06, 0xe9, 0xe0, 0x0e, 0xf4, 0xe0, 0x07, 0x28, 0x18, 0x64, 0xe7, 0x08, 0x40, 0x61, 0x0a,
|
||||
0x00, 0x00, 0x00, 0x00, 0x13, 0x14, 0x72, 0xc0, 0x87, 0x74, 0x60, 0x87, 0x36, 0x68, 0x87, 0x79, 0x68, 0x03, 0x72, 0xc0, 0x87, 0x0d, 0xaf, 0x50, 0x0e, 0x6d, 0xd0, 0x0e, 0x7a, 0x50, 0x0e, 0x6d,
|
||||
0x00, 0x0f, 0x7a, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e, 0x71, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e, 0x78, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e,
|
||||
0x71, 0x60, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xd0, 0x06, 0xe9, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e, 0x76, 0x40, 0x07, 0x7a, 0x60, 0x07, 0x74, 0xd0, 0x06, 0xe6, 0x10,
|
||||
0x07, 0x76, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x60, 0x0e, 0x73, 0x20, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xd0, 0x06, 0xe6, 0x60, 0x07, 0x74, 0xa0, 0x07, 0x76, 0x40, 0x07, 0x6d, 0xe0, 0x0e, 0x78,
|
||||
0xa0, 0x07, 0x71, 0x60, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x76, 0x40, 0x07, 0x43, 0x9e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x86, 0x3c, 0x04, 0x10, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0c, 0x79, 0x0e, 0x20, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x18, 0xf2, 0x10, 0x40, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x30, 0xe4, 0x61, 0x80, 0x00, 0x18, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x60, 0xc8, 0xe3, 0x00, 0x01, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xc0, 0x90, 0x27, 0x02,
|
||||
0x02, 0x60, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x80, 0x2c, 0x10, 0x0a, 0x00, 0x00, 0x00, 0x32, 0x1e, 0x98, 0x14, 0x19, 0x11, 0x4c, 0x90, 0x8c, 0x09, 0x26, 0x47, 0xc6, 0x04, 0x43, 0x1a,
|
||||
0x25, 0x30, 0x02, 0x50, 0x08, 0xc5, 0x50, 0x18, 0x05, 0x42, 0x6b, 0x04, 0x80, 0x70, 0x81, 0x02, 0x02, 0xd1, 0x9d, 0x01, 0xa0, 0x3a, 0x03, 0x00, 0x79, 0x18, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00,
|
||||
0x1a, 0x03, 0x4c, 0x90, 0x46, 0x02, 0x13, 0xc4, 0x88, 0x0c, 0x6f, 0xec, 0xed, 0x4d, 0x0c, 0x44, 0x06, 0x26, 0x26, 0xc7, 0x05, 0xa6, 0xc6, 0x05, 0x06, 0x66, 0x43, 0x10, 0x4c, 0x10, 0x86, 0x63,
|
||||
0x82, 0x30, 0x20, 0x1b, 0x84, 0x81, 0x98, 0x20, 0x0c, 0xc9, 0x06, 0x61, 0x30, 0x28, 0x8c, 0xcd, 0x4d, 0x10, 0x06, 0x65, 0xc3, 0x80, 0x24, 0xc4, 0x04, 0xa1, 0x82, 0x08, 0x4c, 0x10, 0x86, 0x65,
|
||||
0x43, 0x42, 0x2c, 0x0c, 0x41, 0x0c, 0x0d, 0x71, 0x6c, 0x08, 0x9c, 0x09, 0xc2, 0xf5, 0x4c, 0x10, 0x96, 0x66, 0xc3, 0x42, 0x40, 0x0c, 0x41, 0x0c, 0x4d, 0x14, 0x45, 0xc7, 0x86, 0x40, 0xda, 0x40,
|
||||
0x3c, 0x13, 0x00, 0x4c, 0x10, 0x04, 0x60, 0x03, 0xb0, 0x61, 0x20, 0x2c, 0x6b, 0x43, 0x70, 0x6d, 0x18, 0x86, 0x0a, 0x23, 0xd1, 0x16, 0x96, 0xe6, 0x36, 0x41, 0xc0, 0x9c, 0x09, 0xc2, 0xc0, 0x6c,
|
||||
0x18, 0xb8, 0x61, 0xd8, 0x40, 0x10, 0x9b, 0xd1, 0x6d, 0x28, 0x2a, 0x0d, 0xa0, 0xbc, 0x2a, 0x6c, 0x6c, 0x76, 0x6d, 0x2e, 0x69, 0x64, 0x65, 0x6e, 0x74, 0x53, 0x82, 0xa0, 0x0a, 0x19, 0x9e, 0x8b,
|
||||
0x5d, 0x99, 0xdc, 0x5c, 0xda, 0x9b, 0xdb, 0x94, 0x80, 0x68, 0x42, 0x86, 0xe7, 0x62, 0x17, 0xc6, 0x66, 0x57, 0x26, 0x37, 0x25, 0x30, 0xea, 0x90, 0xe1, 0xb9, 0xcc, 0xa1, 0x85, 0x91, 0x95, 0xc9,
|
||||
0x35, 0xbd, 0x91, 0x95, 0xb1, 0x4d, 0x09, 0x92, 0x32, 0x64, 0x78, 0x2e, 0x72, 0x65, 0x73, 0x6f, 0x75, 0x72, 0x63, 0x65, 0x73, 0x53, 0x82, 0xa9, 0x12, 0x19, 0x9e, 0x0b, 0x5d, 0x1e, 0x5c, 0x59,
|
||||
0x90, 0x9b, 0xdb, 0x1b, 0x5d, 0x18, 0x5d, 0xda, 0x9b, 0xdb, 0xdc, 0x94, 0x00, 0xab, 0x43, 0x86, 0xe7, 0x52, 0xe6, 0x46, 0x27, 0x97, 0x07, 0xf5, 0x96, 0xe6, 0x46, 0x37, 0x37, 0x25, 0xf0, 0x00,
|
||||
0x79, 0x18, 0x00, 0x00, 0x4c, 0x00, 0x00, 0x00, 0x33, 0x08, 0x80, 0x1c, 0xc4, 0xe1, 0x1c, 0x66, 0x14, 0x01, 0x3d, 0x88, 0x43, 0x38, 0x84, 0xc3, 0x8c, 0x42, 0x80, 0x07, 0x79, 0x78, 0x07, 0x73,
|
||||
0x98, 0x71, 0x0c, 0xe6, 0x00, 0x0f, 0xed, 0x10, 0x0e, 0xf4, 0x80, 0x0e, 0x33, 0x0c, 0x42, 0x1e, 0xc2, 0xc1, 0x1d, 0xce, 0xa1, 0x1c, 0x66, 0x30, 0x05, 0x3d, 0x88, 0x43, 0x38, 0x84, 0x83, 0x1b,
|
||||
0xcc, 0x03, 0x3d, 0xc8, 0x43, 0x3d, 0x8c, 0x03, 0x3d, 0xcc, 0x78, 0x8c, 0x74, 0x70, 0x07, 0x7b, 0x08, 0x07, 0x79, 0x48, 0x87, 0x70, 0x70, 0x07, 0x7a, 0x70, 0x03, 0x76, 0x78, 0x87, 0x70, 0x20,
|
||||
0x87, 0x19, 0xcc, 0x11, 0x0e, 0xec, 0x90, 0x0e, 0xe1, 0x30, 0x0f, 0x6e, 0x30, 0x0f, 0xe3, 0xf0, 0x0e, 0xf0, 0x50, 0x0e, 0x33, 0x10, 0xc4, 0x1d, 0xde, 0x21, 0x1c, 0xd8, 0x21, 0x1d, 0xc2, 0x61,
|
||||
0x1e, 0x66, 0x30, 0x89, 0x3b, 0xbc, 0x83, 0x3b, 0xd0, 0x43, 0x39, 0xb4, 0x03, 0x3c, 0xbc, 0x83, 0x3c, 0x84, 0x03, 0x3b, 0xcc, 0xf0, 0x14, 0x76, 0x60, 0x07, 0x7b, 0x68, 0x07, 0x37, 0x68, 0x87,
|
||||
0x72, 0x68, 0x07, 0x37, 0x80, 0x87, 0x70, 0x90, 0x87, 0x70, 0x60, 0x07, 0x76, 0x28, 0x07, 0x76, 0xf8, 0x05, 0x76, 0x78, 0x87, 0x77, 0x80, 0x87, 0x5f, 0x08, 0x87, 0x71, 0x18, 0x87, 0x72, 0x98,
|
||||
0x87, 0x79, 0x98, 0x81, 0x2c, 0xee, 0xf0, 0x0e, 0xee, 0xe0, 0x0e, 0xf5, 0xc0, 0x0e, 0xec, 0x30, 0x03, 0x62, 0xc8, 0xa1, 0x1c, 0xe4, 0xa1, 0x1c, 0xcc, 0xa1, 0x1c, 0xe4, 0xa1, 0x1c, 0xdc, 0x61,
|
||||
0x1c, 0xca, 0x21, 0x1c, 0xc4, 0x81, 0x1d, 0xca, 0x61, 0x06, 0xd6, 0x90, 0x43, 0x39, 0xc8, 0x43, 0x39, 0x98, 0x43, 0x39, 0xc8, 0x43, 0x39, 0xb8, 0xc3, 0x38, 0x94, 0x43, 0x38, 0x88, 0x03, 0x3b,
|
||||
0x94, 0xc3, 0x2f, 0xbc, 0x83, 0x3c, 0xfc, 0x82, 0x3b, 0xd4, 0x03, 0x3b, 0xb0, 0xc3, 0x8c, 0xcc, 0x21, 0x07, 0x7c, 0x70, 0x03, 0x74, 0x60, 0x07, 0x37, 0x90, 0x87, 0x72, 0x98, 0x87, 0x77, 0xa8,
|
||||
0x07, 0x79, 0x18, 0x87, 0x72, 0x70, 0x83, 0x70, 0xa0, 0x07, 0x7a, 0x90, 0x87, 0x74, 0x10, 0x87, 0x7a, 0xa0, 0x87, 0x72, 0x00, 0x00, 0x00, 0x00, 0x71, 0x20, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
|
||||
0x66, 0x40, 0x0d, 0x97, 0xef, 0x3c, 0x7e, 0x40, 0x15, 0x05, 0x11, 0x95, 0x0e, 0x30, 0xf8, 0xc8, 0x6d, 0x5b, 0x41, 0x35, 0x5c, 0xbe, 0xf3, 0xf8, 0x01, 0x55, 0x14, 0x44, 0xc4, 0x4e, 0x4e, 0x44,
|
||||
0xf8, 0xc8, 0x6d, 0x1b, 0x81, 0x34, 0x5c, 0xbe, 0xf3, 0xf8, 0x42, 0x44, 0x00, 0x13, 0x11, 0x02, 0xcd, 0xb0, 0x10, 0x16, 0x20, 0x0d, 0x97, 0xef, 0x3c, 0xfe, 0x74, 0x44, 0x04, 0x30, 0x88, 0x83,
|
||||
0x8f, 0xdc, 0xb6, 0x09, 0x58, 0xc3, 0xe5, 0x3b, 0x8f, 0x6f, 0x01, 0x15, 0xa1, 0x09, 0x13, 0x52, 0x11, 0xe8, 0xe3, 0x23, 0xb7, 0x6d, 0x03, 0xdb, 0x70, 0xf9, 0xce, 0xe3, 0x5b, 0x40, 0x45, 0xac,
|
||||
0x04, 0x30, 0x94, 0x40, 0x43, 0x7c, 0x48, 0x24, 0x4d, 0x3e, 0x72, 0xdb, 0x06, 0x40, 0x30, 0x00, 0xd2, 0x00, 0x00, 0x00, 0x61, 0x20, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x13, 0x04, 0x43, 0x2c,
|
||||
0x10, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x34, 0x66, 0x00, 0x4a, 0xae, 0x74, 0x03, 0x0a, 0x31, 0xa0, 0xd8, 0x03, 0xca, 0x3d, 0xa0, 0x14, 0x03, 0xc8, 0x94, 0xc0, 0x08, 0x00, 0x00, 0x00,
|
||||
0x23, 0x06, 0x09, 0x00, 0x82, 0x60, 0xc0, 0x58, 0x88, 0x20, 0x49, 0xcd, 0x88, 0x41, 0x02, 0x80, 0x20, 0x18, 0x30, 0x57, 0x22, 0x4c, 0x93, 0x33, 0x62, 0x60, 0x00, 0x20, 0x08, 0x06, 0xc4, 0x96,
|
||||
0x50, 0x23, 0x06, 0x08, 0x00, 0x82, 0x60, 0x10, 0x5d, 0x89, 0x10, 0x54, 0xa3, 0x09, 0x01, 0x30, 0x4b, 0x10, 0x8c, 0x18, 0x18, 0x00, 0x08, 0x82, 0x01, 0xc1, 0x29, 0xc1, 0x70, 0x83, 0x10, 0x80,
|
||||
0xc1, 0x2c, 0x83, 0x10, 0x04, 0x23, 0x06, 0x08, 0x00, 0x82, 0x60, 0x70, 0x7c, 0xcb, 0x80, 0x24, 0x23, 0x06, 0x0d, 0x00, 0x82, 0x60, 0xe0, 0x74, 0x0b, 0x62, 0x68, 0x41, 0x14, 0x45, 0x0a, 0x02,
|
||||
0x00, 0x00, 0x00, 0x00,
|
||||
};
|
||||
static const D3D12_SHADER_BYTECODE reconvergence_code = SHADER_BYTECODE(reconvergence_dxil);
|
||||
|
||||
/* Compiled with Version: dxcompiler.dll: 1.6 - 1.6.2112.12 (770ac0cc1); dxil.dll: 1.6(101.6.2112.2)
|
||||
* This version of DXC seems to workaround the control flow issue and explicitly hoists the wave-op out of the branch
|
||||
* to avoid convergence, most curious! */
|
||||
#if 0
|
||||
@dx.break.cond = internal constant [1 x i32] zeroinitializer
|
||||
define void @main() {
|
||||
%1 = load i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @dx.break.cond, i32 0, i32 0)
|
||||
%2 = icmp eq i32 %1, 0
|
||||
%3 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 0, i32 0, i1 false) ; CreateHandle(resourceClass,rangeId,index,nonUniformIndex)
|
||||
%4 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 0, i32 0, i32 0, i1 false) ; CreateHandle(resourceClass,rangeId,index,nonUniformIndex)
|
||||
%5 = call i32 @dx.op.threadId.i32(i32 93, i32 0) ; ThreadId(component)
|
||||
%6 = call %dx.types.ResRet.i32 @dx.op.bufferLoad.i32(i32 68, %dx.types.Handle %4, i32 %5, i32 0) ; BufferLoad(srv,index,wot)
|
||||
%7 = extractvalue %dx.types.ResRet.i32 %6, 0
|
||||
br label %8
|
||||
|
||||
; <label>:8 ; preds = %13, %0
|
||||
%9 = call i32 @dx.op.waveReadLaneFirst.i32(i32 118, i32 %7) ; WaveReadLaneFirst(value)
|
||||
%10 = icmp eq i32 %7, %9
|
||||
br i1 %10, label %11, label %13
|
||||
|
||||
; <label>:11 ; preds = %8
|
||||
%12 = call i32 @dx.op.waveActiveOp.i32(i32 119, i32 %7, i8 0, i8 1) ; WaveActiveOp(value,op,sop)
|
||||
; Wow ... Load a constant 0 from a Private array and branch on that.
|
||||
br i1 %2, label %14, label %13
|
||||
|
||||
; <label>:13 ; preds = %11, %8
|
||||
br label %8
|
||||
|
||||
; <label>:14 ; preds = %11
|
||||
call void @dx.op.bufferStore.i32(i32 69, %dx.types.Handle %3, i32 %5, i32 0, i32 %12, i32 undef, i32 undef, i32 undef, i8 1) ; BufferStore(uav,coord0,coord1,value0,value1,value2,value3,mask)
|
||||
ret void
|
||||
}
|
||||
#endif
|
||||
static const BYTE nonconvergence_dxil[] =
|
||||
{
|
||||
0x44, 0x58, 0x42, 0x43, 0x6a, 0xe3, 0x1a, 0x21, 0xec, 0x33, 0x8f, 0x47, 0xcb, 0xd9, 0x75, 0x47, 0x59, 0x62, 0x3f, 0x1f, 0x01, 0x00, 0x00, 0x00, 0x6c, 0x07, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x38, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x58, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00, 0xe8, 0x00, 0x00, 0x00, 0x04, 0x01, 0x00, 0x00, 0x53, 0x46, 0x49, 0x30, 0x08, 0x00, 0x00, 0x00,
|
||||
0x00, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x49, 0x53, 0x47, 0x31, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x4f, 0x53, 0x47, 0x31, 0x08, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x50, 0x53, 0x56, 0x30, 0x78, 0x00, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff, 0x05, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x41, 0x53, 0x48, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x96, 0xd4, 0x61, 0x94, 0x9f, 0xd5, 0x2a, 0x70, 0x1d, 0xed, 0x1f, 0xc3,
|
||||
0x53, 0x48, 0x48, 0x9e, 0x44, 0x58, 0x49, 0x4c, 0x60, 0x06, 0x00, 0x00, 0x60, 0x00, 0x05, 0x00, 0x98, 0x01, 0x00, 0x00, 0x44, 0x58, 0x49, 0x4c, 0x00, 0x01, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00,
|
||||
0x48, 0x06, 0x00, 0x00, 0x42, 0x43, 0xc0, 0xde, 0x21, 0x0c, 0x00, 0x00, 0x8f, 0x01, 0x00, 0x00, 0x0b, 0x82, 0x20, 0x00, 0x02, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x07, 0x81, 0x23, 0x91,
|
||||
0x41, 0xc8, 0x04, 0x49, 0x06, 0x10, 0x32, 0x39, 0x92, 0x01, 0x84, 0x0c, 0x25, 0x05, 0x08, 0x19, 0x1e, 0x04, 0x8b, 0x62, 0x80, 0x14, 0x45, 0x02, 0x42, 0x92, 0x0b, 0x42, 0xa4, 0x10, 0x32, 0x14,
|
||||
0x38, 0x08, 0x18, 0x4b, 0x0a, 0x32, 0x52, 0x88, 0x48, 0x90, 0x14, 0x20, 0x43, 0x46, 0x88, 0xa5, 0x00, 0x19, 0x32, 0x42, 0xe4, 0x48, 0x0e, 0x90, 0x91, 0x22, 0xc4, 0x50, 0x41, 0x51, 0x81, 0x8c,
|
||||
0xe1, 0x83, 0xe5, 0x8a, 0x04, 0x29, 0x46, 0x06, 0x51, 0x18, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x1b, 0x8c, 0xe0, 0xff, 0xff, 0xff, 0xff, 0x07, 0x40, 0x02, 0xa8, 0x0d, 0x84, 0xf0, 0xff, 0xff,
|
||||
0xff, 0xff, 0x03, 0x20, 0x6d, 0x30, 0x86, 0xff, 0xff, 0xff, 0xff, 0x1f, 0x00, 0x09, 0xa8, 0x00, 0x49, 0x18, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x13, 0x82, 0x60, 0x42, 0x20, 0x4c, 0x08, 0x06,
|
||||
0x00, 0x00, 0x00, 0x00, 0x89, 0x20, 0x00, 0x00, 0x3e, 0x00, 0x00, 0x00, 0x32, 0x22, 0x48, 0x09, 0x20, 0x64, 0x85, 0x04, 0x93, 0x22, 0xa4, 0x84, 0x04, 0x93, 0x22, 0xe3, 0x84, 0xa1, 0x90, 0x14,
|
||||
0x12, 0x4c, 0x8a, 0x8c, 0x0b, 0x84, 0xa4, 0x4c, 0x10, 0x74, 0x73, 0x04, 0x60, 0x90, 0x01, 0x80, 0xc2, 0x08, 0x40, 0x09, 0x06, 0x91, 0x32, 0x00, 0x00, 0xc8, 0xcc, 0x11, 0x20, 0xa5, 0x00, 0x00,
|
||||
0x20, 0x44, 0x89, 0xd0, 0x3d, 0xc3, 0xe5, 0x4f, 0xd8, 0x43, 0x48, 0x7e, 0x08, 0x34, 0xc3, 0x42, 0xa0, 0x60, 0xcd, 0x11, 0x04, 0xc5, 0x60, 0x00, 0x01, 0xd0, 0xc8, 0x15, 0x65, 0x00, 0x06, 0x00,
|
||||
0x00, 0x00, 0x20, 0x82, 0x37, 0x0d, 0x97, 0x3f, 0x61, 0x0f, 0x21, 0xf9, 0x2b, 0x21, 0xad, 0xc4, 0xe4, 0x23, 0xb7, 0x8d, 0x0a, 0x00, 0x00, 0x00, 0xa5, 0x90, 0x80, 0x01, 0x40, 0x73, 0x20, 0x60,
|
||||
0x26, 0x32, 0x18, 0x07, 0x76, 0x08, 0x87, 0x79, 0x98, 0x07, 0x37, 0x98, 0x05, 0x7a, 0x90, 0x87, 0x7a, 0x18, 0x07, 0x7a, 0xa8, 0x07, 0x79, 0x28, 0x07, 0x72, 0x10, 0x85, 0x7a, 0x30, 0x07, 0x73,
|
||||
0x28, 0x07, 0x79, 0xe0, 0x83, 0x7a, 0x70, 0x87, 0x79, 0x48, 0x87, 0x73, 0x70, 0x87, 0x72, 0x20, 0x07, 0x30, 0x48, 0x07, 0x77, 0xa0, 0x07, 0x3f, 0x40, 0x01, 0x20, 0x3b, 0x93, 0x19, 0x8c, 0x03,
|
||||
0x3b, 0x84, 0xc3, 0x3c, 0xcc, 0x83, 0x1b, 0xc8, 0xc2, 0x2d, 0xcc, 0x02, 0x3d, 0xc8, 0x43, 0x3d, 0x8c, 0x03, 0x3d, 0xd4, 0x83, 0x3c, 0x94, 0x03, 0x39, 0x88, 0x42, 0x3d, 0x98, 0x83, 0x39, 0x94,
|
||||
0x83, 0x3c, 0xf0, 0x41, 0x3d, 0xb8, 0xc3, 0x3c, 0xa4, 0xc3, 0x39, 0xb8, 0x43, 0x39, 0x90, 0x03, 0x18, 0xa4, 0x83, 0x3b, 0xd0, 0x83, 0x1f, 0xa0, 0x00, 0x10, 0x9e, 0x23, 0x00, 0x05, 0x02, 0x53,
|
||||
0x00, 0x00, 0x00, 0x00, 0x13, 0x14, 0x72, 0xc0, 0x87, 0x74, 0x60, 0x87, 0x36, 0x68, 0x87, 0x79, 0x68, 0x03, 0x72, 0xc0, 0x87, 0x0d, 0xaf, 0x50, 0x0e, 0x6d, 0xd0, 0x0e, 0x7a, 0x50, 0x0e, 0x6d,
|
||||
0x00, 0x0f, 0x7a, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e, 0x71, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e, 0x78, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e,
|
||||
0x71, 0x60, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xd0, 0x06, 0xe9, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e, 0x76, 0x40, 0x07, 0x7a, 0x60, 0x07, 0x74, 0xd0, 0x06, 0xe6, 0x10,
|
||||
0x07, 0x76, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x60, 0x0e, 0x73, 0x20, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xd0, 0x06, 0xe6, 0x60, 0x07, 0x74, 0xa0, 0x07, 0x76, 0x40, 0x07, 0x6d, 0xe0, 0x0e, 0x78,
|
||||
0xa0, 0x07, 0x71, 0x60, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x76, 0x40, 0x07, 0x3a, 0x0f, 0x24, 0x90, 0x21, 0x23, 0x25, 0x40, 0x00, 0x1e, 0xa6, 0x31, 0xe4, 0x21, 0x00, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x60, 0xc8, 0x63, 0x00, 0x01, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xc0, 0x90, 0x27, 0x01, 0x02, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x80, 0x21, 0x8f, 0x01, 0x04, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x43, 0x1e, 0x07, 0x08, 0x80, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x86, 0x3c, 0x10, 0x10, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0c, 0x79, 0x26, 0x20, 0x00, 0x06, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xc8, 0x02, 0x01, 0x0b, 0x00, 0x00, 0x00, 0x32, 0x1e, 0x98, 0x14,
|
||||
0x19, 0x11, 0x4c, 0x90, 0x8c, 0x09, 0x26, 0x47, 0xc6, 0x04, 0x43, 0x02, 0x25, 0x30, 0x02, 0x50, 0x0c, 0x85, 0x51, 0x08, 0x05, 0x42, 0xba, 0x40, 0x01, 0x81, 0xa8, 0x8d, 0x00, 0xd0, 0x9d, 0x01,
|
||||
0xa0, 0x3c, 0x03, 0x40, 0x61, 0x04, 0x00, 0x00, 0x79, 0x18, 0x00, 0x00, 0x3f, 0x00, 0x00, 0x00, 0x1a, 0x03, 0x4c, 0x90, 0x46, 0x02, 0x13, 0x44, 0x35, 0x18, 0x63, 0x0b, 0x73, 0x3b, 0x03, 0xb1,
|
||||
0x2b, 0x93, 0x9b, 0x4b, 0x7b, 0x73, 0x03, 0x99, 0x71, 0xb9, 0x01, 0x41, 0xa1, 0x0b, 0x3b, 0x9b, 0x7b, 0x91, 0x2a, 0x62, 0x2a, 0x0a, 0x9a, 0x2a, 0xfa, 0x9a, 0xb9, 0x81, 0x79, 0x31, 0x4b, 0x73,
|
||||
0x0b, 0x63, 0x4b, 0xd9, 0x10, 0x04, 0x13, 0x04, 0x00, 0x99, 0x20, 0x00, 0xc9, 0x06, 0x61, 0x20, 0x26, 0x08, 0x80, 0xb2, 0x41, 0x18, 0x0c, 0x0a, 0x63, 0x73, 0x1b, 0x06, 0xc4, 0x20, 0x26, 0x08,
|
||||
0x17, 0x44, 0x60, 0x82, 0x00, 0x2c, 0x13, 0x04, 0x80, 0xd9, 0x20, 0x0c, 0xcd, 0x86, 0x84, 0x50, 0x16, 0x82, 0x18, 0x18, 0xc2, 0xd9, 0x10, 0x3c, 0x13, 0x84, 0x2c, 0x9a, 0x20, 0x34, 0xcf, 0x86,
|
||||
0x85, 0x88, 0x16, 0x82, 0x18, 0x18, 0x49, 0x92, 0x9c, 0x0d, 0xc1, 0xb4, 0x81, 0x80, 0x28, 0x00, 0x98, 0x20, 0x14, 0x01, 0x89, 0xb6, 0xb0, 0x34, 0xb7, 0x09, 0x82, 0xe6, 0x4c, 0x10, 0x80, 0x66,
|
||||
0xc3, 0x90, 0x0d, 0xc3, 0x06, 0x82, 0xc0, 0x1a, 0x6d, 0x43, 0x61, 0x5d, 0x40, 0xb5, 0x55, 0x61, 0x63, 0xb3, 0x6b, 0x73, 0x49, 0x23, 0x2b, 0x73, 0xa3, 0x9b, 0x12, 0x04, 0x55, 0xc8, 0xf0, 0x5c,
|
||||
0xec, 0xca, 0xe4, 0xe6, 0xd2, 0xde, 0xdc, 0xa6, 0x04, 0x44, 0x13, 0x32, 0x3c, 0x17, 0xbb, 0x30, 0x36, 0xbb, 0x32, 0xb9, 0x29, 0x81, 0x51, 0x87, 0x0c, 0xcf, 0x65, 0x0e, 0x2d, 0x8c, 0xac, 0x4c,
|
||||
0xae, 0xe9, 0x8d, 0xac, 0x8c, 0x6d, 0x4a, 0x80, 0x94, 0x21, 0xc3, 0x73, 0x91, 0x2b, 0x9b, 0x7b, 0xab, 0x93, 0x1b, 0x2b, 0x9b, 0x9b, 0x12, 0x50, 0x75, 0xc8, 0xf0, 0x5c, 0xca, 0xdc, 0xe8, 0xe4,
|
||||
0xf2, 0xa0, 0xde, 0xd2, 0xdc, 0xe8, 0xe6, 0xa6, 0x04, 0x1b, 0x00, 0x00, 0x79, 0x18, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00, 0x33, 0x08, 0x80, 0x1c, 0xc4, 0xe1, 0x1c, 0x66, 0x14, 0x01, 0x3d, 0x88,
|
||||
0x43, 0x38, 0x84, 0xc3, 0x8c, 0x42, 0x80, 0x07, 0x79, 0x78, 0x07, 0x73, 0x98, 0x71, 0x0c, 0xe6, 0x00, 0x0f, 0xed, 0x10, 0x0e, 0xf4, 0x80, 0x0e, 0x33, 0x0c, 0x42, 0x1e, 0xc2, 0xc1, 0x1d, 0xce,
|
||||
0xa1, 0x1c, 0x66, 0x30, 0x05, 0x3d, 0x88, 0x43, 0x38, 0x84, 0x83, 0x1b, 0xcc, 0x03, 0x3d, 0xc8, 0x43, 0x3d, 0x8c, 0x03, 0x3d, 0xcc, 0x78, 0x8c, 0x74, 0x70, 0x07, 0x7b, 0x08, 0x07, 0x79, 0x48,
|
||||
0x87, 0x70, 0x70, 0x07, 0x7a, 0x70, 0x03, 0x76, 0x78, 0x87, 0x70, 0x20, 0x87, 0x19, 0xcc, 0x11, 0x0e, 0xec, 0x90, 0x0e, 0xe1, 0x30, 0x0f, 0x6e, 0x30, 0x0f, 0xe3, 0xf0, 0x0e, 0xf0, 0x50, 0x0e,
|
||||
0x33, 0x10, 0xc4, 0x1d, 0xde, 0x21, 0x1c, 0xd8, 0x21, 0x1d, 0xc2, 0x61, 0x1e, 0x66, 0x30, 0x89, 0x3b, 0xbc, 0x83, 0x3b, 0xd0, 0x43, 0x39, 0xb4, 0x03, 0x3c, 0xbc, 0x83, 0x3c, 0x84, 0x03, 0x3b,
|
||||
0xcc, 0xf0, 0x14, 0x76, 0x60, 0x07, 0x7b, 0x68, 0x07, 0x37, 0x68, 0x87, 0x72, 0x68, 0x07, 0x37, 0x80, 0x87, 0x70, 0x90, 0x87, 0x70, 0x60, 0x07, 0x76, 0x28, 0x07, 0x76, 0xf8, 0x05, 0x76, 0x78,
|
||||
0x87, 0x77, 0x80, 0x87, 0x5f, 0x08, 0x87, 0x71, 0x18, 0x87, 0x72, 0x98, 0x87, 0x79, 0x98, 0x81, 0x2c, 0xee, 0xf0, 0x0e, 0xee, 0xe0, 0x0e, 0xf5, 0xc0, 0x0e, 0xec, 0x30, 0x03, 0x62, 0xc8, 0xa1,
|
||||
0x1c, 0xe4, 0xa1, 0x1c, 0xcc, 0xa1, 0x1c, 0xe4, 0xa1, 0x1c, 0xdc, 0x61, 0x1c, 0xca, 0x21, 0x1c, 0xc4, 0x81, 0x1d, 0xca, 0x61, 0x06, 0xd6, 0x90, 0x43, 0x39, 0xc8, 0x43, 0x39, 0x98, 0x43, 0x39,
|
||||
0xc8, 0x43, 0x39, 0xb8, 0xc3, 0x38, 0x94, 0x43, 0x38, 0x88, 0x03, 0x3b, 0x94, 0xc3, 0x2f, 0xbc, 0x83, 0x3c, 0xfc, 0x82, 0x3b, 0xd4, 0x03, 0x3b, 0xb0, 0xc3, 0x0c, 0xc4, 0x21, 0x07, 0x7c, 0x70,
|
||||
0x03, 0x7a, 0x28, 0x87, 0x76, 0x80, 0x87, 0x19, 0xcc, 0x43, 0x0e, 0xf8, 0xe0, 0x06, 0xe2, 0x20, 0x0f, 0xe5, 0x10, 0x0e, 0xeb, 0xe0, 0x06, 0xe2, 0x20, 0x0f, 0x33, 0x22, 0x88, 0x1c, 0xf0, 0xc1,
|
||||
0x0d, 0xc8, 0x41, 0x1c, 0xce, 0xc1, 0x0d, 0xec, 0x21, 0x1c, 0xe4, 0x81, 0x1d, 0xc2, 0x21, 0x1f, 0xde, 0xa1, 0x1e, 0xe8, 0x01, 0x00, 0x00, 0x00, 0x71, 0x20, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
|
||||
0x06, 0xd0, 0x0c, 0x97, 0x1f, 0x44, 0x04, 0xa0, 0xf8, 0x82, 0xd3, 0x0c, 0x76, 0x40, 0x0d, 0x97, 0xef, 0x3c, 0x7e, 0x40, 0x15, 0x05, 0x11, 0x95, 0x0e, 0x30, 0xf8, 0xc8, 0x6d, 0x9b, 0x41, 0x35,
|
||||
0x5c, 0xbe, 0xf3, 0xf8, 0x01, 0x55, 0x14, 0x44, 0xc4, 0x4e, 0x4e, 0x44, 0xf8, 0xc8, 0x6d, 0x5b, 0x81, 0x34, 0x5c, 0xbe, 0xf3, 0xf8, 0x42, 0x44, 0x00, 0x13, 0x11, 0x02, 0xcd, 0xb0, 0x10, 0x26,
|
||||
0x20, 0x0d, 0x97, 0xef, 0x3c, 0xfe, 0x74, 0x44, 0x04, 0x30, 0x88, 0x83, 0x8f, 0xdc, 0xb6, 0x0d, 0x58, 0xc3, 0xe5, 0x3b, 0x8f, 0x6f, 0x01, 0x15, 0xa1, 0x09, 0x13, 0x52, 0x11, 0xe8, 0xe3, 0x23,
|
||||
0xb7, 0x6d, 0x04, 0xdb, 0x70, 0xf9, 0xce, 0xe3, 0x5b, 0x40, 0x45, 0xac, 0x04, 0x30, 0x94, 0x40, 0x43, 0x7c, 0x48, 0x24, 0x4d, 0x3e, 0x72, 0xdb, 0x16, 0x40, 0x30, 0x00, 0xd2, 0x00, 0x00, 0x00,
|
||||
0x61, 0x20, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x13, 0x04, 0x45, 0x2c, 0x10, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x04, 0x66, 0x00, 0x4a, 0xae, 0x74, 0x03, 0x0a, 0x31, 0xa0, 0xd8, 0x03,
|
||||
0xca, 0x3d, 0xa0, 0x14, 0x03, 0x08, 0x95, 0xc0, 0x08, 0x00, 0xed, 0xa1, 0x8e, 0x40, 0x00, 0x80, 0x04, 0x48, 0x00, 0x00, 0x14, 0x00, 0x30, 0xdc, 0x10, 0x54, 0x60, 0x30, 0x62, 0x90, 0x00, 0x20,
|
||||
0x08, 0x06, 0x8e, 0xb6, 0x14, 0x96, 0x05, 0x8d, 0x18, 0x24, 0x00, 0x08, 0x82, 0x81, 0xb3, 0x31, 0xc5, 0x75, 0x45, 0x23, 0x06, 0x06, 0x00, 0x82, 0x60, 0x60, 0x7c, 0x0c, 0x36, 0x62, 0x80, 0x00,
|
||||
0x20, 0x08, 0x06, 0xd3, 0xc6, 0x08, 0x41, 0x36, 0x9a, 0x10, 0x00, 0xb3, 0x04, 0xc1, 0x88, 0x81, 0x01, 0x80, 0x20, 0x18, 0x18, 0x60, 0xd0, 0x04, 0xc3, 0x0d, 0x42, 0x00, 0x06, 0xb3, 0x0c, 0xc2,
|
||||
0x10, 0x8c, 0x18, 0x20, 0x00, 0x08, 0x82, 0x41, 0x32, 0x06, 0xce, 0xb0, 0x30, 0xb3, 0x0c, 0xc4, 0x90, 0xcc, 0x12, 0x04, 0x23, 0x06, 0x0d, 0x00, 0x82, 0x60, 0x00, 0x85, 0x81, 0x83, 0x18, 0x5e,
|
||||
0x40, 0x51, 0x54, 0x83, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
};
|
||||
static const D3D12_SHADER_BYTECODE nonconvergence_code = SHADER_BYTECODE(nonconvergence_dxil);
|
||||
|
||||
memset(root_parameters, 0, sizeof(root_parameters));
|
||||
memset(&rs_desc, 0, sizeof(rs_desc));
|
||||
root_parameters[0].ParameterType = D3D12_ROOT_PARAMETER_TYPE_SRV;
|
||||
root_parameters[0].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
|
||||
root_parameters[1].ParameterType = D3D12_ROOT_PARAMETER_TYPE_UAV;
|
||||
root_parameters[1].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
|
||||
rs_desc.pParameters = root_parameters;
|
||||
rs_desc.NumParameters = ARRAY_SIZE(root_parameters);
|
||||
|
||||
if (!init_compute_test_context(&context))
|
||||
return;
|
||||
|
||||
if (!context_supports_dxil(&context))
|
||||
{
|
||||
skip("Context does not support DXIL.\n");
|
||||
destroy_test_context(&context);
|
||||
return;
|
||||
}
|
||||
|
||||
create_root_signature(context.device, &rs_desc, &context.root_signature);
|
||||
pso_converged = create_compute_pipeline_state(context.device, context.root_signature, reconvergence_code);
|
||||
pso_nonconverged = create_compute_pipeline_state(context.device, context.root_signature, nonconvergence_code);
|
||||
|
||||
inputs = create_upload_buffer(context.device, sizeof(inputs_data), inputs_data);
|
||||
outputs = create_default_buffer(context.device, sizeof(inputs_data) * 2,
|
||||
D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
|
||||
|
||||
ID3D12GraphicsCommandList_SetComputeRootSignature(context.list, context.root_signature);
|
||||
ID3D12GraphicsCommandList_SetPipelineState(context.list, pso_converged);
|
||||
ID3D12GraphicsCommandList_SetComputeRootShaderResourceView(context.list, 0,
|
||||
ID3D12Resource_GetGPUVirtualAddress(inputs));
|
||||
ID3D12GraphicsCommandList_SetComputeRootUnorderedAccessView(context.list, 1,
|
||||
ID3D12Resource_GetGPUVirtualAddress(outputs));
|
||||
ID3D12GraphicsCommandList_Dispatch(context.list, 1, 1, 1);
|
||||
ID3D12GraphicsCommandList_SetComputeRootUnorderedAccessView(context.list, 1,
|
||||
ID3D12Resource_GetGPUVirtualAddress(outputs) + sizeof(reference_converged));
|
||||
ID3D12GraphicsCommandList_SetPipelineState(context.list, pso_nonconverged);
|
||||
ID3D12GraphicsCommandList_Dispatch(context.list, 1, 1, 1);
|
||||
|
||||
transition_resource_state(context.list, outputs,
|
||||
D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE);
|
||||
|
||||
get_buffer_readback_with_command_list(outputs, DXGI_FORMAT_R32_UINT, &rb, context.queue, context.list);
|
||||
|
||||
for (i = 0; i < ARRAY_SIZE(reference_converged); i++)
|
||||
{
|
||||
uint32_t v = get_readback_uint(&rb, i, 0, 0);
|
||||
ok(v == reference_converged[i], "Element %u, %u != %u.\n", i, v, reference_converged[i]);
|
||||
}
|
||||
|
||||
for (i = 0; i < ARRAY_SIZE(reference_nonconverged); i++)
|
||||
{
|
||||
uint32_t v = get_readback_uint(&rb, i + 16, 0, 0);
|
||||
ok(v == reference_nonconverged[i], "Element %u, %u != %u.\n", i, v, reference_nonconverged[i]);
|
||||
}
|
||||
|
||||
release_resource_readback(&rb);
|
||||
|
||||
ID3D12Resource_Release(inputs);
|
||||
ID3D12Resource_Release(outputs);
|
||||
ID3D12PipelineState_Release(pso_converged);
|
||||
ID3D12PipelineState_Release(pso_nonconverged);
|
||||
destroy_test_context(&context);
|
||||
}
|
||||
|
||||
void test_shader_sm65_wave_intrinsics(void)
|
||||
{
|
||||
D3D12_FEATURE_DATA_SHADER_MODEL shader_model;
|
||||
|
|
|
|||
|
|
@ -303,3 +303,4 @@ decl_test(test_mesh_shader_rendering);
|
|||
decl_test(test_mesh_shader_execute_indirect);
|
||||
decl_test(test_amplification_shader);
|
||||
decl_test(test_advanced_cbv_layout);
|
||||
decl_test(test_shader_waveop_maximal_convergence);
|
||||
|
|
|
|||
Loading…
Reference in New Issue