projects / platform / upstream / VK-GL-CTS.git / commitdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: 44e1fdd)
Fix array stride in uniform buffer for VK_KHR_16bit_storage
authorLei Zhang <antiagainst@google.com>
Fri, 14 Sep 2018 18:56:12 +0000 (14:56 -0400)
committerAlexander Galazin <Alexander.Galazin@arm.com>
Tue, 25 Sep 2018 18:09:15 +0000 (14:09 -0400)
Array stride need to be a multiple of 16.

Affects:
dEQP-VK.spirv_assembly.instruction.*.16bit_storage.*

Components: Vulkan, SPIR-V
VK-GL-CTS issue: 1344

Change-Id: If5c4415e8f1914a2579d914d33ab3b334affd8d3

external/vulkancts/modules/vulkan/spirv_assembly/vktSpvAsm16bitStorageTests.cpp patch | blob | history

diff --git a/external/vulkancts/modules/vulkan/spirv_assembly/vktSpvAsm16bitStorageTests.cpp b/external/vulkancts/modules/vulkan/spirv_assembly/vktSpvAsm16bitStorageTests.cpp
index bab7d48..b99d05d 100644 (file)
--- a/external/vulkancts/modules/vulkan/spirv_assembly/vktSpvAsm16bitStorageTests.cpp
+++ b/external/vulkancts/modules/vulkan/spirv_assembly/vktSpvAsm16bitStorageTests.cpp
@@ -226,7 +226,7 @@ bool graphicsCheck16BitFloats64 (const std::vector<Resource>&       originalFloats,
        for (deUint32 outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
        {
                vector<deUint8> originalBytes;
-               originalFloats[outputNdx].getBytes(originalBytes);
+               originalFloats[outputNdx].getBuffer()->getPackedBytes(originalBytes);
 
                const deUint16* returned        = static_cast<const deUint16*>(outputAllocs[outputNdx]->getHostPtr());
                const double*   original        = reinterpret_cast<const double*>(&originalBytes.front());
@@ -304,7 +304,7 @@ bool computeCheck16BitFloats64 (const std::vector<Resource>&        originalFloats,
        for (deUint32 outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
        {
                vector<deUint8> originalBytes;
-               originalFloats[outputNdx].getBytes(originalBytes);
+               originalFloats[outputNdx].getBuffer()->getPackedBytes(originalBytes);
 
                const deUint16* returned        = static_cast<const deUint16*>(outputAllocs[outputNdx]->getHostPtr());
                const double*   original        = reinterpret_cast<const double*>(&originalBytes.front());
@@ -333,7 +333,7 @@ bool check64BitFloats (const std::vector<Resource>&         /* originalFloats */,
        for (deUint32 outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
        {
                vector<deUint8> expectedBytes;
-               expectedOutputs[outputNdx].getBytes(expectedBytes);
+               expectedOutputs[outputNdx].getBuffer()->getPackedBytes(expectedBytes);
 
                const double*   returnedAsDouble        = static_cast<const double*>(outputAllocs[outputNdx]->getHostPtr());
                const double*   expectedAsDouble        = reinterpret_cast<const double*>(&expectedBytes.front());
@@ -7092,7 +7092,7 @@ void addGraphics16BitStorageUniformFloat16To64Group (tcu::TestCaseGroup* testGro
 
                const StringTemplate decoration         (
                        "OpDecorate %ra_f64 ArrayStride 8\n"
-                       "OpDecorate %ra_f16 ArrayStride 2\n"
+                       "OpDecorate %ra_f16 ArrayStride ${stride16}\n"
                        "OpMemberDecorate %SSBO64 0 Offset 0\n"
                        "OpMemberDecorate %SSBO16 0 Offset 0\n"
                        "OpDecorate %SSBO64 BufferBlock\n"
@@ -7147,10 +7147,13 @@ void addGraphics16BitStorageUniformFloat16To64Group (tcu::TestCaseGroup* testGro
                                string                          testName        = string(CAPABILITIES[capIdx].name) + "_scalar_float";
                                bool                            useConstIdx     = constantIndices[constIndexIdx].useConstantIndex;
                                deUint32                        constIdx        = constantIndices[constIndexIdx].constantIndex;
+                               const bool                      isUBO           = CAPABILITIES[capIdx].dtype == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
 
                                specs["cap"]                                    = CAPABILITIES[capIdx].cap;
                                specs["indecor"]                                = CAPABILITIES[capIdx].decor;
                                specs["constarrayidx"]                  = de::toString(constIdx);
+                               specs["stride16"]                               = isUBO ? "16" : "2";
+
                                if (useConstIdx)
                                        specs["arrayindex"] = "c_i32_ci";
                                else
@@ -7166,7 +7169,7 @@ void addGraphics16BitStorageUniformFloat16To64Group (tcu::TestCaseGroup* testGro
                                for (deUint32 numIdx = 0; numIdx < numDataPoints; ++numIdx)
                                        float64Data.push_back(deFloat16To64(float16Data[useConstIdx ? constIdx : numIdx]));
 
-                               resources.inputs.push_back(Resource(BufferSp(new Float16Buffer(float16Data)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
+                               resources.inputs.push_back(Resource(BufferSp(new Float16Buffer(float16Data, isUBO ? 14 : 0)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
                                resources.outputs.push_back(Resource(BufferSp(new Float64Buffer(float64Data)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
                                resources.verifyIO = check64BitFloats;
                                resources.inputs.back().setDescriptorType(CAPABILITIES[capIdx].dtype);
@@ -7202,7 +7205,7 @@ void addGraphics16BitStorageUniformFloat16To64Group (tcu::TestCaseGroup* testGro
 
                const StringTemplate decoration         (
                        "OpDecorate %ra_v2f64 ArrayStride 16\n"
-                       "OpDecorate %ra_v2f16 ArrayStride 4\n"
+                       "OpDecorate %ra_v2f16 ArrayStride ${stride16}\n"
                        "OpMemberDecorate %SSBO64 0 Offset 0\n"
                        "OpMemberDecorate %SSBO16 0 Offset 0\n"
                        "OpDecorate %SSBO64 BufferBlock\n"
@@ -7257,10 +7260,13 @@ void addGraphics16BitStorageUniformFloat16To64Group (tcu::TestCaseGroup* testGro
                                string                          testName        = string(CAPABILITIES[capIdx].name) + "_vector_float";
                                bool                            useConstIdx     = constantIndices[constIndexIdx].useConstantIndex;
                                deUint32                        constIdx        = constantIndices[constIndexIdx].constantIndex;
+                               const bool                      isUBO           = CAPABILITIES[capIdx].dtype == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
 
                                specs["cap"]                                    = CAPABILITIES[capIdx].cap;
                                specs["indecor"]                                = CAPABILITIES[capIdx].decor;
                                specs["constarrayidx"]                  = de::toString(constIdx);
+                               specs["stride16"]                               = isUBO ? "16" : "4";
+
                                if (useConstIdx)
                                        specs["arrayindex"] = "c_i32_ci";
                                else
@@ -7276,7 +7282,7 @@ void addGraphics16BitStorageUniformFloat16To64Group (tcu::TestCaseGroup* testGro
                                for (deUint32 numIdx = 0; numIdx < numDataPoints; ++numIdx)
                                        float64Data.push_back(deFloat16To64(float16Data[constantIndices[constIndexIdx].useConstantIndex ? (constantIndices[constIndexIdx].constantIndex * 2 + numIdx % 2) : numIdx]));
 
-                               resources.inputs.push_back(Resource(BufferSp(new Float16Buffer(float16Data)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
+                               resources.inputs.push_back(Resource(BufferSp(new Float16Buffer(float16Data, isUBO ? 12 : 0)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
                                resources.outputs.push_back(Resource(BufferSp(new Float64Buffer(float64Data)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
                                resources.verifyIO = check64BitFloats;
                                resources.inputs.back().setDescriptorType(CAPABILITIES[capIdx].dtype);
@@ -7634,7 +7640,7 @@ void addCompute16bitStorageUniform64To16Group (tcu::TestCaseGroup* group)
                "OpExecutionMode %main LocalSize 1 1 1\n"
                "OpDecorate %id BuiltIn GlobalInvocationId\n"
 
-               "${stride}"
+               "${stride}\n"
 
                "OpMemberDecorate %SSBO64 0 Offset 0\n"
                "OpMemberDecorate %SSBO16 0 Offset 0\n"
@@ -7728,15 +7734,19 @@ void addCompute16bitStorageUniform64To16Group (tcu::TestCaseGroup* group)
                        const char*     name;
                        const char*     base64;
                        const char*     base16;
-                       const char*     stride;
+                       const char*     strideStr;
+                       const char* stride64UBO;
+                       unsigned        padding64UBO;
+                       const char* stride64SSBO;
+                       unsigned        padding64SSBO;
                        unsigned        count;
                };
 
                const CompositeType     cTypes[]        =
                {
-                       {"scalar",      "f64",          "f16",          "OpDecorate %f64arr ArrayStride 8\nOpDecorate %f16arr ArrayStride 2\n",                         numElements},
-                       {"vector",      "v4f64",        "v4f16",        "OpDecorate %v4f64arr ArrayStride 32\nOpDecorate %v4f16arr ArrayStride 8\n",            numElements / 4},
-                       {"matrix",      "v4f64",        "v4f16",        "OpDecorate %m2v4f64arr ArrayStride 64\nOpDecorate %m2v4f16arr ArrayStride 16\n",       numElements / 8},
+                       {"scalar",      "f64",          "f16",          "OpDecorate %f16arr ArrayStride 2\nOpDecorate %f64arr ArrayStride ",                    "16",   8,      "8",    0,      numElements},
+                       {"vector",      "v4f64",        "v4f16",        "OpDecorate %v4f16arr ArrayStride 8\nOpDecorate %v4f64arr ArrayStride ",                "32",   0,      "32",   0,      numElements / 4},
+                       {"matrix",      "v4f64",        "v4f16",        "OpDecorate %m2v4f16arr ArrayStride 16\nOpDecorate %m2v4f64arr ArrayStride ",   "64",   0,      "64",   0,      numElements / 8},
                };
 
                vector<double>          float64Data                     = getFloat64s(rnd, numElements);
@@ -7749,16 +7759,22 @@ void addCompute16bitStorageUniform64To16Group (tcu::TestCaseGroup* group)
                                        ComputeShaderSpec               spec;
                                        map<string, string>             specs;
                                        string                                  testName        = string(CAPABILITIES[capIdx].name) + "_" + cTypes[tyIdx].name + "_float_" + rndModes[rndModeIdx].name;
+                                       const bool                              isUBO           = CAPABILITIES[capIdx].dtype == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
 
                                        specs["capability"]             = CAPABILITIES[capIdx].cap;
                                        specs["storage"]                = CAPABILITIES[capIdx].decor;
-                                       specs["stride"]                 = cTypes[tyIdx].stride;
+                                       specs["stride"]                 = cTypes[tyIdx].strideStr;
                                        specs["base64"]                 = cTypes[tyIdx].base64;
                                        specs["base16"]                 = cTypes[tyIdx].base16;
                                        specs["rounding"]               = rndModes[rndModeIdx].decor;
                                        specs["types"]                  = floatTypes;
                                        specs["convert"]                = "OpFConvert";
 
+                                       if (isUBO)
+                                               specs["stride"] += cTypes[tyIdx].stride64UBO;
+                                       else
+                                               specs["stride"] += cTypes[tyIdx].stride64SSBO;
+
                                        if (deStringEqual(cTypes[tyIdx].name, "matrix"))
                                        {
                                                if (strcmp(rndModes[rndModeIdx].name, "rtz") == 0)
@@ -7789,8 +7805,9 @@ void addCompute16bitStorageUniform64To16Group (tcu::TestCaseGroup* group)
                                        spec.assembly                   = shaderTemplate.specialize(specs);
                                        spec.numWorkGroups              = IVec3(cTypes[tyIdx].count, 1, 1);
                                        spec.verifyIO                   = rndModes[rndModeIdx].func;
+                                       const unsigned padding  = isUBO ? cTypes[tyIdx].padding64UBO : cTypes[tyIdx].padding64SSBO;
 
-                                       spec.inputs.push_back(Resource(BufferSp(new Float64Buffer(float64Data)), CAPABILITIES[capIdx].dtype));
+                                       spec.inputs.push_back(Resource(BufferSp(new Float64Buffer(float64Data, padding)), CAPABILITIES[capIdx].dtype));
                                        // We provided a custom verifyIO in the above in which inputs will be used for checking.
                                        // So put dummy data in the expected values.
                                        spec.outputs.push_back(BufferSp(new Float16Buffer(float16DummyData)));
@@ -7817,7 +7834,6 @@ void addGraphics16BitStorageUniformFloat64To16Group (tcu::TestCaseGroup* testGro
                                                                                                                         "OpCapability Float16\n"
                                                                                                                         "OpCapability ${cap}\n");
 
-       resources.inputs.push_back(Resource(BufferSp(new Float64Buffer(float64Data)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
        // We use a custom verifyIO to check the result via computing directly from inputs; the contents in outputs do not matter.
        resources.outputs.push_back(Resource(BufferSp(new Float16Buffer(float16DummyData)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
 
@@ -7852,7 +7868,7 @@ void addGraphics16BitStorageUniformFloat64To16Group (tcu::TestCaseGroup* testGro
                        "   %ssbo16 = OpVariable %up_SSBO16 Uniform\n";
 
                const StringTemplate decoration         (
-                       "OpDecorate %ra_f64 ArrayStride 8\n"
+                       "OpDecorate %ra_f64 ArrayStride ${stride64}\n"
                        "OpDecorate %ra_f16 ArrayStride 2\n"
                        "OpMemberDecorate %SSBO64 0 Offset 0\n"
                        "OpMemberDecorate %SSBO16 0 Offset 0\n"
@@ -7911,15 +7927,19 @@ void addGraphics16BitStorageUniformFloat64To16Group (tcu::TestCaseGroup* testGro
                        {
                                map<string, string>     specs;
                                string                          testName        = string(CAPABILITIES[capIdx].name) + "_scalar_float_" + rndModes[rndModeIdx].name;
+                               const bool                      isUBO           = CAPABILITIES[capIdx].dtype == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
 
                                specs["cap"]                                    = CAPABILITIES[capIdx].cap;
                                specs["indecor"]                                = CAPABILITIES[capIdx].decor;
                                specs["rounddecor"]                             = rndModes[rndModeIdx].decor;
+                               specs["stride64"]                               = isUBO ? "16" : "8";
 
                                fragments["capability"]                 = capabilities.specialize(specs);
                                fragments["decoration"]                 = decoration.specialize(specs);
 
-                               resources.inputs.back().setDescriptorType(CAPABILITIES[capIdx].dtype);
+                               resources.inputs.clear();
+                               resources.inputs.push_back(Resource(BufferSp(new Float64Buffer(float64Data, isUBO ? 8 : 0)), CAPABILITIES[capIdx].dtype));
+
                                resources.verifyIO                              = rndModes[rndModeIdx].f;
                                VulkanFeatures features = get16BitStorageFeatures(CAPABILITIES[capIdx].name);
                                features.coreFeatures.shaderFloat64 = DE_TRUE;
@@ -8015,7 +8035,8 @@ void addGraphics16BitStorageUniformFloat64To16Group (tcu::TestCaseGroup* testGro
                                fragments["capability"]                 = capabilities.specialize(specs);
                                fragments["decoration"]                 = decoration.specialize(specs);
 
-                               resources.inputs.back().setDescriptorType(CAPABILITIES[capIdx].dtype);
+                               resources.inputs.clear();
+                               resources.inputs.push_back(Resource(BufferSp(new Float64Buffer(float64Data)), CAPABILITIES[capIdx].dtype));
                                resources.verifyIO                              = rndModes[rndModeIdx].f;
                                VulkanFeatures features = get16BitStorageFeatures(CAPABILITIES[capIdx].name);
                                features.coreFeatures.shaderFloat64 = DE_TRUE;
@@ -8131,7 +8152,8 @@ void addGraphics16BitStorageUniformFloat64To16Group (tcu::TestCaseGroup* testGro
                                fragments["capability"]                 = capabilities.specialize(specs);
                                fragments["decoration"]                 = decoration.specialize(specs);
 
-                               resources.inputs.back().setDescriptorType(CAPABILITIES[capIdx].dtype);
+                               resources.inputs.clear();
+                               resources.inputs.push_back(Resource(BufferSp(new Float64Buffer(float64Data)), CAPABILITIES[capIdx].dtype));
                                resources.verifyIO                              = rndModes[rndModeIdx].f;
                                VulkanFeatures features = get16BitStorageFeatures(CAPABILITIES[capIdx].name);
                                features.coreFeatures.shaderFloat64 = DE_TRUE;
@@ -8298,7 +8320,7 @@ void addCompute16bitStorageUniform16To64Group (tcu::TestCaseGroup* group)
                "OpExecutionMode %main LocalSize 1 1 1\n"
                "OpDecorate %id BuiltIn GlobalInvocationId\n"
 
-               "${stride}"
+               "${stride}\n"
 
                "OpMemberDecorate %SSBO64 0 Offset 0\n"
                "OpMemberDecorate %SSBO16 0 Offset 0\n"
@@ -8377,7 +8399,11 @@ void addCompute16bitStorageUniform16To64Group (tcu::TestCaseGroup* group)
                        const char*     name;
                        const char*     base64;
                        const char*     base16;
-                       const char*     stride;
+                       const char*     strideStr;
+                       const char* stride16UBO;
+                       unsigned        padding16UBO;
+                       const char* stride16SSBO;
+                       unsigned        padding16SSBO;
                        bool            useConstantIndex;
                        unsigned        constantIndex;
                        unsigned        count;
@@ -8385,11 +8411,11 @@ void addCompute16bitStorageUniform16To64Group (tcu::TestCaseGroup* group)
 
                const CompositeType     cTypes[]        =
                {
-                       {"scalar",                              "f64",          "f16",          "OpDecorate %f64arr ArrayStride 8\nOpDecorate %f16arr ArrayStride 2\n",                         false,  0,      numElements},
-                       {"scalar_const_idx_5",  "f64",          "f16",          "OpDecorate %f64arr ArrayStride 8\nOpDecorate %f16arr ArrayStride 2\n",                         true,   5,      numElements},
-                       {"scalar_const_idx_8",  "f64",          "f16",          "OpDecorate %f64arr ArrayStride 8\nOpDecorate %f16arr ArrayStride 2\n",                         true,   8,      numElements},
-                       {"vector",                              "v2f64",        "v2f16",        "OpDecorate %v2f64arr ArrayStride 16\nOpDecorate %v2f16arr ArrayStride 4\n",            false,  0,      numElements / 2},
-                       {"matrix",                              "v2f64",        "v2f16",        "OpDecorate %m4v2f64arr ArrayStride 64\nOpDecorate %m4v2f16arr ArrayStride 16\n",       false,  0,      numElements / 8},
+                       {"scalar",                              "f64",          "f16",          "OpDecorate %f64arr ArrayStride 8\nOpDecorate %f16arr ArrayStride ",                    "16",   14,     "2",    0,      false,  0,      numElements},
+                       {"scalar_const_idx_5",  "f64",          "f16",          "OpDecorate %f64arr ArrayStride 8\nOpDecorate %f16arr ArrayStride ",                    "16",   14,     "2",    0,      true,   5,      numElements},
+                       {"scalar_const_idx_8",  "f64",          "f16",          "OpDecorate %f64arr ArrayStride 8\nOpDecorate %f16arr ArrayStride ",                    "16",   14,     "2",    0,      true,   8,      numElements},
+                       {"vector",                              "v2f64",        "v2f16",        "OpDecorate %v2f64arr ArrayStride 16\nOpDecorate %v2f16arr ArrayStride ",               "16",   12,     "4",    0,      false,  0,      numElements / 2},
+                       {"matrix",                              "v2f64",        "v2f16",        "OpDecorate %m4v2f64arr ArrayStride 64\nOpDecorate %m4v2f16arr ArrayStride ",   "16",   0, "16",        0,      false,  0,      numElements / 8},
                };
 
                vector<deFloat16>       float16Data                     = getFloat16s(rnd, numElements);
@@ -8405,15 +8431,22 @@ void addCompute16bitStorageUniform16To64Group (tcu::TestCaseGroup* group)
                                ComputeShaderSpec               spec;
                                map<string, string>             specs;
                                string                                  testName        = string(CAPABILITIES[capIdx].name) + "_" + cTypes[tyIdx].name + "_float";
+                               const bool                              isUBO           = CAPABILITIES[capIdx].dtype == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
 
                                specs["capability"]             = CAPABILITIES[capIdx].cap;
                                specs["storage"]                = CAPABILITIES[capIdx].decor;
-                               specs["stride"]                 = cTypes[tyIdx].stride;
+                               specs["stride"]                 = cTypes[tyIdx].strideStr;
                                specs["base64"]                 = cTypes[tyIdx].base64;
                                specs["base16"]                 = cTypes[tyIdx].base16;
                                specs["types"]                  = floatTypes;
                                specs["convert"]                = "OpFConvert";
                                specs["constarrayidx"]  = de::toString(cTypes[tyIdx].constantIndex);
+
+                               if (isUBO)
+                                       specs["stride"] += cTypes[tyIdx].stride16UBO;
+                               else
+                                       specs["stride"] += cTypes[tyIdx].stride16SSBO;
+
                                if (cTypes[tyIdx].useConstantIndex)
                                        specs["arrayindex"] = "c_i32_ci";
                                else
@@ -8464,8 +8497,9 @@ void addCompute16bitStorageUniform16To64Group (tcu::TestCaseGroup* group)
                                spec.assembly                   = shaderTemplate.specialize(specs);
                                spec.numWorkGroups              = IVec3(cTypes[tyIdx].count, 1, 1);
                                spec.verifyIO                   = check64BitFloats;
+                               const unsigned padding  = isUBO ? cTypes[tyIdx].padding16UBO : cTypes[tyIdx].padding16SSBO;
 
-                               spec.inputs.push_back(Resource(BufferSp(new Float16Buffer(float16Data)), CAPABILITIES[capIdx].dtype));
+                               spec.inputs.push_back(Resource(BufferSp(new Float16Buffer(float16Data, padding)), CAPABILITIES[capIdx].dtype));
                                spec.outputs.push_back(Resource(BufferSp(new Float64Buffer(cTypes[tyIdx].useConstantIndex ? float64DataConstIdx : float64Data))));
                                spec.extensions.push_back("VK_KHR_16bit_storage");
                                spec.extensions.push_back("VK_KHR_shader_float16_int8");
Domain: Graphics System / GL;