/*-------------------------------------------------------------------------
* drawElements Quality Program OpenGL ES 3.0 Module
* -------------------------------------------------
*
* Copyright 2014 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*//*!
* \file
* \brief Mipmapping tests.
*//*--------------------------------------------------------------------*/
#include "es3fTextureMipmapTests.hpp"
#include "glsTextureTestUtil.hpp"
#include "gluTexture.hpp"
#include "gluTextureUtil.hpp"
#include "gluPixelTransfer.hpp"
#include "tcuTextureUtil.hpp"
#include "tcuMatrix.hpp"
#include "tcuMatrixUtil.hpp"
#include "tcuTexLookupVerifier.hpp"
#include "tcuVectorUtil.hpp"
#include "deStringUtil.hpp"
#include "deRandom.hpp"
#include "deString.h"
#include "glwFunctions.hpp"
#include "glwEnums.hpp"
using std::vector;
using std::string;
using namespace deqp::gls;
namespace deqp
{
namespace gles3
{
namespace Functional
{
using std::string;
using std::vector;
using tcu::TestLog;
using tcu::Vec2;
using tcu::Vec3;
using tcu::Vec4;
using tcu::IVec4;
using namespace gls::TextureTestUtil;
static float getMinLodForCell (int cellNdx)
{
static const float s_values[] =
{
1.0f,
3.5f,
2.0f,
-2.0f,
0.0f,
3.0f,
10.0f,
4.8f,
5.8f,
5.7f,
-1.9f,
4.0f,
6.5f,
7.1f,
-1e10,
1000.f
};
return s_values[cellNdx % DE_LENGTH_OF_ARRAY(s_values)];
}
static float getMaxLodForCell (int cellNdx)
{
static const float s_values[] =
{
0.0f,
0.2f,
0.7f,
0.4f,
1.3f,
0.0f,
0.5f,
1.2f,
-2.0f,
1.0f,
0.1f,
0.3f,
2.7f,
1.2f,
10.0f,
-1000.f,
1e10f
};
return s_values[cellNdx % DE_LENGTH_OF_ARRAY(s_values)];
}
enum CoordType
{
COORDTYPE_BASIC, //!< texCoord = translateScale(position).
COORDTYPE_BASIC_BIAS, //!< Like basic, but with bias values.
COORDTYPE_AFFINE, //!< texCoord = translateScaleRotateShear(position).
COORDTYPE_PROJECTED, //!< Projected coordinates, w != 1
COORDTYPE_LAST
};
// Texture2DMipmapCase
class Texture2DMipmapCase : public tcu::TestCase
{
public:
Texture2DMipmapCase (tcu::TestContext& testCtx,
glu::RenderContext& renderCtx,
const glu::ContextInfo& renderCtxInfo,
const char* name,
const char* desc,
CoordType coordType,
deUint32 minFilter,
deUint32 wrapS,
deUint32 wrapT,
deUint32 format,
deUint32 dataType,
int width,
int height);
~Texture2DMipmapCase (void);
void init (void);
void deinit (void);
IterateResult iterate (void);
private:
Texture2DMipmapCase (const Texture2DMipmapCase& other);
Texture2DMipmapCase& operator= (const Texture2DMipmapCase& other);
glu::RenderContext& m_renderCtx;
const glu::ContextInfo& m_renderCtxInfo;
CoordType m_coordType;
deUint32 m_minFilter;
deUint32 m_wrapS;
deUint32 m_wrapT;
deUint32 m_format;
deUint32 m_dataType;
int m_width;
int m_height;
glu::Texture2D* m_texture;
TextureRenderer m_renderer;
};
Texture2DMipmapCase::Texture2DMipmapCase (tcu::TestContext& testCtx,
glu::RenderContext& renderCtx,
const glu::ContextInfo& renderCtxInfo,
const char* name,
const char* desc,
CoordType coordType,
deUint32 minFilter,
deUint32 wrapS,
deUint32 wrapT,
deUint32 format,
deUint32 dataType,
int width,
int height)
: TestCase (testCtx, name, desc)
, m_renderCtx (renderCtx)
, m_renderCtxInfo (renderCtxInfo)
, m_coordType (coordType)
, m_minFilter (minFilter)
, m_wrapS (wrapS)
, m_wrapT (wrapT)
, m_format (format)
, m_dataType (dataType)
, m_width (width)
, m_height (height)
, m_texture (DE_NULL)
, m_renderer (renderCtx, testCtx.getLog(), glu::GLSL_VERSION_300_ES, glu::PRECISION_HIGHP)
{
}
Texture2DMipmapCase::~Texture2DMipmapCase (void)
{
deinit();
}
void Texture2DMipmapCase::init (void)
{
if (m_coordType == COORDTYPE_PROJECTED && m_renderCtx.getRenderTarget().getNumSamples() > 0)
throw tcu::NotSupportedError("Projected lookup validation not supported in multisample config");
m_texture = new glu::Texture2D(m_renderCtx, m_format, m_dataType, m_width, m_height);
int numLevels = deLog2Floor32(de::max(m_width, m_height))+1;
// Fill texture with colored grid.
for (int levelNdx = 0; levelNdx < numLevels; levelNdx++)
{
deUint32 step = 0xff / (numLevels-1);
deUint32 inc = deClamp32(step*levelNdx, 0x00, 0xff);
deUint32 dec = 0xff - inc;
deUint32 rgb = (inc << 16) | (dec << 8) | 0xff;
deUint32 color = 0xff000000 | rgb;
m_texture->getRefTexture().allocLevel(levelNdx);
tcu::clear(m_texture->getRefTexture().getLevel(levelNdx), toVec4(tcu::RGBA(color)));
}
}
void Texture2DMipmapCase::deinit (void)
{
delete m_texture;
m_texture = DE_NULL;
m_renderer.clear();
}
static void getBasicTexCoord2D (std::vector<float>& dst, int cellNdx)
{
static const struct
{
Vec2 bottomLeft;
Vec2 topRight;
} s_basicCoords[] =
{
{ Vec2(-0.1f, 0.1f), Vec2( 0.8f, 1.0f) },
{ Vec2(-0.3f, -0.6f), Vec2( 0.7f, 0.4f) },
{ Vec2(-0.3f, 0.6f), Vec2( 0.7f, -0.9f) },
{ Vec2(-0.8f, 0.6f), Vec2( 0.7f, -0.9f) },
{ Vec2(-0.5f, -0.5f), Vec2( 1.5f, 1.5f) },
{ Vec2( 1.0f, -1.0f), Vec2(-1.3f, 1.0f) },
{ Vec2( 1.2f, -1.0f), Vec2(-1.3f, 1.6f) },
{ Vec2( 2.2f, -1.1f), Vec2(-1.3f, 0.8f) },
{ Vec2(-1.5f, 1.6f), Vec2( 1.7f, -1.4f) },
{ Vec2( 2.0f, 1.6f), Vec2( 2.3f, -1.4f) },
{ Vec2( 1.3f, -2.6f), Vec2(-2.7f, 2.9f) },
{ Vec2(-0.8f, -6.6f), Vec2( 6.0f, -0.9f) },
{ Vec2( -8.0f, 9.0f), Vec2( 8.3f, -7.0f) },
{ Vec2(-16.0f, 10.0f), Vec2( 18.3f, 24.0f) },
{ Vec2( 30.2f, 55.0f), Vec2(-24.3f, -1.6f) },
{ Vec2(-33.2f, 64.1f), Vec2( 32.1f, -64.1f) },
};
DE_ASSERT(de::inBounds(cellNdx, 0, DE_LENGTH_OF_ARRAY(s_basicCoords)));
const Vec2& bottomLeft = s_basicCoords[cellNdx].bottomLeft;
const Vec2& topRight = s_basicCoords[cellNdx].topRight;
computeQuadTexCoord2D(dst, bottomLeft, topRight);
}
static void getAffineTexCoord2D (std::vector<float>& dst, int cellNdx)
{
// Use basic coords as base.
getBasicTexCoord2D(dst, cellNdx);
// Rotate based on cell index.
float angle = 2.0f*DE_PI * ((float)cellNdx / 16.0f);
tcu::Mat2 rotMatrix = tcu::rotationMatrix(angle);
// Second and third row are sheared.
float shearX = de::inRange(cellNdx, 4, 11) ? (float)(15-cellNdx) / 16.0f : 0.0f;
tcu::Mat2 shearMatrix = tcu::shearMatrix(tcu::Vec2(shearX, 0.0f));
tcu::Mat2 transform = rotMatrix * shearMatrix;
Vec2 p0 = transform * Vec2(dst[0], dst[1]);
Vec2 p1 = transform * Vec2(dst[2], dst[3]);
Vec2 p2 = transform * Vec2(dst[4], dst[5]);
Vec2 p3 = transform * Vec2(dst[6], dst[7]);
dst[0] = p0.x(); dst[1] = p0.y();
dst[2] = p1.x(); dst[3] = p1.y();
dst[4] = p2.x(); dst[5] = p2.y();
dst[6] = p3.x(); dst[7] = p3.y();
}
Texture2DMipmapCase::IterateResult Texture2DMipmapCase::iterate (void)
{
const glw::Functions& gl = m_renderCtx.getFunctions();
const tcu::Texture2D& refTexture = m_texture->getRefTexture();
const deUint32 magFilter = GL_NEAREST;
const int texWidth = refTexture.getWidth();
const int texHeight = refTexture.getHeight();
const int defViewportWidth = texWidth*4;
const int defViewportHeight = texHeight*4;
const RandomViewport viewport (m_renderCtx.getRenderTarget(), defViewportWidth, defViewportHeight, deStringHash(getName()));
ReferenceParams sampleParams (TEXTURETYPE_2D);
vector<float> texCoord;
const bool isProjected = m_coordType == COORDTYPE_PROJECTED;
const bool useLodBias = m_coordType == COORDTYPE_BASIC_BIAS;
tcu::Surface renderedFrame (viewport.width, viewport.height);
// Viewport is divided into 4x4 grid.
int gridWidth = 4;
int gridHeight = 4;
int cellWidth = viewport.width / gridWidth;
int cellHeight = viewport.height / gridHeight;
// Bail out if rendertarget is too small.
if (viewport.width < defViewportWidth/2 || viewport.height < defViewportHeight/2)
throw tcu::NotSupportedError("Too small viewport", "", __FILE__, __LINE__);
// Sampling parameters.
sampleParams.sampler = glu::mapGLSampler(m_wrapS, m_wrapT, m_minFilter, magFilter);
sampleParams.samplerType = gls::TextureTestUtil::getSamplerType(m_texture->getRefTexture().getFormat());
sampleParams.flags = (isProjected ? ReferenceParams::PROJECTED : 0) | (useLodBias ? ReferenceParams::USE_BIAS : 0);
sampleParams.lodMode = LODMODE_EXACT; // Use ideal lod.
// Upload texture data.
m_texture->upload();
// Bind gradient texture and setup sampler parameters.
gl.bindTexture (GL_TEXTURE_2D, m_texture->getGLTexture());
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, m_wrapS);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, m_wrapT);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, m_minFilter);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, magFilter);
GLU_EXPECT_NO_ERROR(gl.getError(), "After texture setup");
// Bias values.
static const float s_bias[] = { 1.0f, -2.0f, 0.8f, -0.5f, 1.5f, 0.9f, 2.0f, 4.0f };
// Projection values.
static const Vec4 s_projections[] =
{
Vec4(1.2f, 1.0f, 0.7f, 1.0f),
Vec4(1.3f, 0.8f, 0.6f, 2.0f),
Vec4(0.8f, 1.0f, 1.7f, 0.6f),
Vec4(1.2f, 1.0f, 1.7f, 1.5f)
};
// Render cells.
for (int gridY = 0; gridY < gridHeight; gridY++)
{
for (int gridX = 0; gridX < gridWidth; gridX++)
{
const int curX = cellWidth*gridX;
const int curY = cellHeight*gridY;
const int curW = gridX+1 == gridWidth ? (viewport.width-curX) : cellWidth;
const int curH = gridY+1 == gridHeight ? (viewport.height-curY) : cellHeight;
const int cellNdx = gridY*gridWidth + gridX;
// Compute texcoord.
switch (m_coordType)
{
case COORDTYPE_BASIC_BIAS: // Fall-through.
case COORDTYPE_PROJECTED:
case COORDTYPE_BASIC: getBasicTexCoord2D (texCoord, cellNdx); break;
case COORDTYPE_AFFINE: getAffineTexCoord2D (texCoord, cellNdx); break;
default: DE_ASSERT(DE_FALSE);
}
if (isProjected)
sampleParams.w = s_projections[cellNdx % DE_LENGTH_OF_ARRAY(s_projections)];
if (useLodBias)
sampleParams.bias = s_bias[cellNdx % DE_LENGTH_OF_ARRAY(s_bias)];
// Render with GL.
gl.viewport(viewport.x+curX, viewport.y+curY, curW, curH);
m_renderer.renderQuad(0, &texCoord[0], sampleParams);
}
}
// Read result.
glu::readPixels(m_renderCtx, viewport.x, viewport.y, renderedFrame.getAccess());
// Compare and log.
{
const tcu::PixelFormat& pixelFormat = m_renderCtx.getRenderTarget().getPixelFormat();
const bool isTrilinear = m_minFilter == GL_NEAREST_MIPMAP_LINEAR || m_minFilter == GL_LINEAR_MIPMAP_LINEAR;
tcu::Surface referenceFrame (viewport.width, viewport.height);
tcu::Surface errorMask (viewport.width, viewport.height);
tcu::LookupPrecision lookupPrec;
tcu::LodPrecision lodPrec;
int numFailedPixels = 0;
lookupPrec.coordBits = tcu::IVec3(20, 20, 0);
lookupPrec.uvwBits = tcu::IVec3(16, 16, 0); // Doesn't really matter since pixels are unicolored.
lookupPrec.colorThreshold = tcu::computeFixedPointThreshold(max(getBitsVec(pixelFormat) - (isTrilinear ? 2 : 1), tcu::IVec4(0)));
lookupPrec.colorMask = getCompareMask(pixelFormat);
lodPrec.derivateBits = 10;
lodPrec.lodBits = isProjected ? 6 : 8;
for (int gridY = 0; gridY < gridHeight; gridY++)
{
for (int gridX = 0; gridX < gridWidth; gridX++)
{
const int curX = cellWidth*gridX;
const int curY = cellHeight*gridY;
const int curW = gridX+1 == gridWidth ? (viewport.width-curX) : cellWidth;
const int curH = gridY+1 == gridHeight ? (viewport.height-curY) : cellHeight;
const int cellNdx = gridY*gridWidth + gridX;
// Compute texcoord.
switch (m_coordType)
{
case COORDTYPE_BASIC_BIAS: // Fall-through.
case COORDTYPE_PROJECTED:
case COORDTYPE_BASIC: getBasicTexCoord2D (texCoord, cellNdx); break;
case COORDTYPE_AFFINE: getAffineTexCoord2D (texCoord, cellNdx); break;
default: DE_ASSERT(DE_FALSE);
}
if (isProjected)
sampleParams.w = s_projections[cellNdx % DE_LENGTH_OF_ARRAY(s_projections)];
if (useLodBias)
sampleParams.bias = s_bias[cellNdx % DE_LENGTH_OF_ARRAY(s_bias)];
// Render ideal result
sampleTexture(SurfaceAccess(referenceFrame, pixelFormat, curX, curY, curW, curH),
refTexture, &texCoord[0], sampleParams);
// Compare this cell
numFailedPixels += computeTextureLookupDiff(tcu::getSubregion(renderedFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(referenceFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(errorMask.getAccess(), curX, curY, curW, curH),
m_texture->getRefTexture(), &texCoord[0], sampleParams,
lookupPrec, lodPrec, m_testCtx.getWatchDog());
}
}
if (numFailedPixels > 0)
m_testCtx.getLog() << TestLog::Message << "ERROR: Image verification failed, found " << numFailedPixels << " invalid pixels!" << TestLog::EndMessage;
m_testCtx.getLog() << TestLog::ImageSet("Result", "Verification result")
<< TestLog::Image("Rendered", "Rendered image", renderedFrame);
if (numFailedPixels > 0)
{
m_testCtx.getLog() << TestLog::Image("Reference", "Ideal reference", referenceFrame)
<< TestLog::Image("ErrorMask", "Error mask", errorMask);
}
m_testCtx.getLog() << TestLog::EndImageSet;
{
const bool isOk = numFailedPixels == 0;
m_testCtx.setTestResult(isOk ? QP_TEST_RESULT_PASS : QP_TEST_RESULT_FAIL,
isOk ? "Pass" : "Image verification failed");
}
}
return STOP;
}
// TextureCubeMipmapCase
class TextureCubeMipmapCase : public tcu::TestCase
{
public:
TextureCubeMipmapCase (tcu::TestContext& testCtx,
glu::RenderContext& renderCtx,
const glu::ContextInfo& renderCtxInfo,
const char* name,
const char* desc,
CoordType coordType,
deUint32 minFilter,
deUint32 wrapS,
deUint32 wrapT,
deUint32 format,
deUint32 dataType,
int size);
~TextureCubeMipmapCase (void);
void init (void);
void deinit (void);
IterateResult iterate (void);
private:
TextureCubeMipmapCase (const TextureCubeMipmapCase& other);
TextureCubeMipmapCase& operator= (const TextureCubeMipmapCase& other);
glu::RenderContext& m_renderCtx;
const glu::ContextInfo& m_renderCtxInfo;
CoordType m_coordType;
deUint32 m_minFilter;
deUint32 m_wrapS;
deUint32 m_wrapT;
deUint32 m_format;
deUint32 m_dataType;
int m_size;
glu::TextureCube* m_texture;
TextureRenderer m_renderer;
};
TextureCubeMipmapCase::TextureCubeMipmapCase (tcu::TestContext& testCtx,
glu::RenderContext& renderCtx,
const glu::ContextInfo& renderCtxInfo,
const char* name,
const char* desc,
CoordType coordType,
deUint32 minFilter,
deUint32 wrapS,
deUint32 wrapT,
deUint32 format,
deUint32 dataType,
int size)
: TestCase (testCtx, name, desc)
, m_renderCtx (renderCtx)
, m_renderCtxInfo (renderCtxInfo)
, m_coordType (coordType)
, m_minFilter (minFilter)
, m_wrapS (wrapS)
, m_wrapT (wrapT)
, m_format (format)
, m_dataType (dataType)
, m_size (size)
, m_texture (DE_NULL)
, m_renderer (renderCtx, testCtx.getLog(), glu::GLSL_VERSION_300_ES, glu::PRECISION_HIGHP)
{
}
TextureCubeMipmapCase::~TextureCubeMipmapCase (void)
{
deinit();
}
void TextureCubeMipmapCase::init (void)
{
if (m_coordType == COORDTYPE_PROJECTED && m_renderCtx.getRenderTarget().getNumSamples() > 0)
throw tcu::NotSupportedError("Projected lookup validation not supported in multisample config");
m_texture = new glu::TextureCube(m_renderCtx, m_format, m_dataType, m_size);
int numLevels = deLog2Floor32(m_size)+1;
// Fill texture with colored grid.
for (int faceNdx = 0; faceNdx < tcu::CUBEFACE_LAST; faceNdx++)
{
for (int levelNdx = 0; levelNdx < numLevels; levelNdx++)
{
deUint32 step = 0xff / (numLevels-1);
deUint32 inc = deClamp32(step*levelNdx, 0x00, 0xff);
deUint32 dec = 0xff - inc;
deUint32 rgb = 0;
switch (faceNdx)
{
case 0: rgb = (inc << 16) | (dec << 8) | 255; break;
case 1: rgb = (255 << 16) | (inc << 8) | dec; break;
case 2: rgb = (dec << 16) | (255 << 8) | inc; break;
case 3: rgb = (dec << 16) | (inc << 8) | 255; break;
case 4: rgb = (255 << 16) | (dec << 8) | inc; break;
case 5: rgb = (inc << 16) | (255 << 8) | dec; break;
}
deUint32 color = 0xff000000 | rgb;
m_texture->getRefTexture().allocLevel((tcu::CubeFace)faceNdx, levelNdx);
tcu::clear(m_texture->getRefTexture().getLevelFace(levelNdx, (tcu::CubeFace)faceNdx), toVec4(tcu::RGBA(color)));
}
}
}
void TextureCubeMipmapCase::deinit (void)
{
delete m_texture;
m_texture = DE_NULL;
m_renderer.clear();
}
static void randomPartition (vector<IVec4>& dst, de::Random& rnd, int x, int y, int width, int height)
{
const int minWidth = 8;
const int minHeight = 8;
bool partition = rnd.getFloat() > 0.4f;
bool partitionX = partition && width > minWidth && rnd.getBool();
bool partitionY = partition && height > minHeight && !partitionX;
if (partitionX)
{
int split = width/2 + rnd.getInt(-width/4, +width/4);
randomPartition(dst, rnd, x, y, split, height);
randomPartition(dst, rnd, x+split, y, width-split, height);
}
else if (partitionY)
{
int split = height/2 + rnd.getInt(-height/4, +height/4);
randomPartition(dst, rnd, x, y, width, split);
randomPartition(dst, rnd, x, y+split, width, height-split);
}
else
dst.push_back(IVec4(x, y, width, height));
}
static void computeGridLayout (vector<IVec4>& dst, int width, int height)
{
de::Random rnd(7);
randomPartition(dst, rnd, 0, 0, width, height);
}
TextureCubeMipmapCase::IterateResult TextureCubeMipmapCase::iterate (void)
{
const deUint32 magFilter = GL_NEAREST;
const int texWidth = m_texture->getRefTexture().getSize();
const int texHeight = m_texture->getRefTexture().getSize();
const int defViewportWidth = texWidth*2;
const int defViewportHeight = texHeight*2;
const glw::Functions& gl = m_renderCtx.getFunctions();
const RandomViewport viewport (m_renderCtx.getRenderTarget(), defViewportWidth, defViewportHeight, deStringHash(getName()));
const bool isProjected = m_coordType == COORDTYPE_PROJECTED;
const bool useLodBias = m_coordType == COORDTYPE_BASIC_BIAS;
vector<float> texCoord;
tcu::Surface renderedFrame (viewport.width, viewport.height);
// Bail out if rendertarget is too small.
if (viewport.width < defViewportWidth/2 || viewport.height < defViewportHeight/2)
throw tcu::NotSupportedError("Too small viewport", "", __FILE__, __LINE__);
// Upload texture data.
m_texture->upload();
// Bind gradient texture and setup sampler parameters.
gl.bindTexture (GL_TEXTURE_CUBE_MAP, m_texture->getGLTexture());
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_S, m_wrapS);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_T, m_wrapT);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MIN_FILTER, m_minFilter);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAG_FILTER, magFilter);
GLU_EXPECT_NO_ERROR(gl.getError(), "After texture setup");
// Compute grid.
vector<IVec4> gridLayout;
computeGridLayout(gridLayout, viewport.width, viewport.height);
// Bias values.
static const float s_bias[] = { 1.0f, -2.0f, 0.8f, -0.5f, 1.5f, 0.9f, 2.0f, 4.0f };
// Projection values \note Less agressive than in 2D case due to smaller quads.
static const Vec4 s_projections[] =
{
Vec4(1.2f, 1.0f, 0.7f, 1.0f),
Vec4(1.3f, 0.8f, 0.6f, 1.1f),
Vec4(0.8f, 1.0f, 1.2f, 0.8f),
Vec4(1.2f, 1.0f, 1.3f, 0.9f)
};
// Render with GL
for (int cellNdx = 0; cellNdx < (int)gridLayout.size(); cellNdx++)
{
const int curX = gridLayout[cellNdx].x();
const int curY = gridLayout[cellNdx].y();
const int curW = gridLayout[cellNdx].z();
const int curH = gridLayout[cellNdx].w();
const tcu::CubeFace cubeFace = (tcu::CubeFace)(cellNdx % tcu::CUBEFACE_LAST);
RenderParams params (TEXTURETYPE_CUBE);
DE_ASSERT(m_coordType != COORDTYPE_AFFINE); // Not supported.
computeQuadTexCoordCube(texCoord, cubeFace);
if (isProjected)
{
params.flags |= ReferenceParams::PROJECTED;
params.w = s_projections[cellNdx % DE_LENGTH_OF_ARRAY(s_projections)];
}
if (useLodBias)
{
params.flags |= ReferenceParams::USE_BIAS;
params.bias = s_bias[cellNdx % DE_LENGTH_OF_ARRAY(s_bias)];
}
// Render with GL.
gl.viewport(viewport.x+curX, viewport.y+curY, curW, curH);
m_renderer.renderQuad(0, &texCoord[0], params);
}
GLU_EXPECT_NO_ERROR(gl.getError(), "Draw");
// Read result.
glu::readPixels(m_renderCtx, viewport.x, viewport.y, renderedFrame.getAccess());
GLU_EXPECT_NO_ERROR(gl.getError(), "Read pixels");
// Render reference and compare
{
tcu::Surface referenceFrame (viewport.width, viewport.height);
tcu::Surface errorMask (viewport.width, viewport.height);
int numFailedPixels = 0;
ReferenceParams params (TEXTURETYPE_CUBE);
tcu::LookupPrecision lookupPrec;
tcu::LodPrecision lodPrec;
// Params for rendering reference
params.sampler = glu::mapGLSampler(m_wrapS, m_wrapT, m_minFilter, magFilter);
params.sampler.seamlessCubeMap = true;
params.lodMode = LODMODE_EXACT;
// Comparison parameters
lookupPrec.colorMask = getCompareMask(m_renderCtx.getRenderTarget().getPixelFormat());
lookupPrec.colorThreshold = tcu::computeFixedPointThreshold(max(getBitsVec(m_renderCtx.getRenderTarget().getPixelFormat())-2, IVec4(0)));
lookupPrec.coordBits = isProjected ? tcu::IVec3(8) : tcu::IVec3(10);
lookupPrec.uvwBits = tcu::IVec3(5,5,0);
lodPrec.derivateBits = 10;
lodPrec.lodBits = isProjected ? 3 : 6;
for (int cellNdx = 0; cellNdx < (int)gridLayout.size(); cellNdx++)
{
const int curX = gridLayout[cellNdx].x();
const int curY = gridLayout[cellNdx].y();
const int curW = gridLayout[cellNdx].z();
const int curH = gridLayout[cellNdx].w();
const tcu::CubeFace cubeFace = (tcu::CubeFace)(cellNdx % tcu::CUBEFACE_LAST);
DE_ASSERT(m_coordType != COORDTYPE_AFFINE); // Not supported.
computeQuadTexCoordCube(texCoord, cubeFace);
if (isProjected)
{
params.flags |= ReferenceParams::PROJECTED;
params.w = s_projections[cellNdx % DE_LENGTH_OF_ARRAY(s_projections)];
}
if (useLodBias)
{
params.flags |= ReferenceParams::USE_BIAS;
params.bias = s_bias[cellNdx % DE_LENGTH_OF_ARRAY(s_bias)];
}
// Render ideal reference.
{
SurfaceAccess idealDst(referenceFrame, m_renderCtx.getRenderTarget().getPixelFormat(), curX, curY, curW, curH);
sampleTexture(idealDst, m_texture->getRefTexture(), &texCoord[0], params);
}
// Compare this cell
numFailedPixels += computeTextureLookupDiff(tcu::getSubregion(renderedFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(referenceFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(errorMask.getAccess(), curX, curY, curW, curH),
m_texture->getRefTexture(), &texCoord[0], params,
lookupPrec, lodPrec, m_testCtx.getWatchDog());
}
if (numFailedPixels > 0)
m_testCtx.getLog() << TestLog::Message << "ERROR: Image verification failed, found " << numFailedPixels << " invalid pixels!" << TestLog::EndMessage;
m_testCtx.getLog() << TestLog::ImageSet("Result", "Verification result")
<< TestLog::Image("Rendered", "Rendered image", renderedFrame);
if (numFailedPixels > 0)
{
m_testCtx.getLog() << TestLog::Image("Reference", "Ideal reference", referenceFrame)
<< TestLog::Image("ErrorMask", "Error mask", errorMask);
}
m_testCtx.getLog() << TestLog::EndImageSet;
{
const bool isOk = numFailedPixels == 0;
m_testCtx.setTestResult(isOk ? QP_TEST_RESULT_PASS : QP_TEST_RESULT_FAIL,
isOk ? "Pass" : "Image verification failed");
}
}
return STOP;
}
// Texture2DGenMipmapCase
class Texture2DGenMipmapCase : public tcu::TestCase
{
public:
Texture2DGenMipmapCase (tcu::TestContext& testCtx, glu::RenderContext& renderCtx, const char* name, const char* desc, deUint32 format, deUint32 dataType, deUint32 hint, int width, int height);
~Texture2DGenMipmapCase (void);
void init (void);
void deinit (void);
IterateResult iterate (void);
private:
Texture2DGenMipmapCase (const Texture2DGenMipmapCase& other);
Texture2DGenMipmapCase& operator= (const Texture2DGenMipmapCase& other);
glu::RenderContext& m_renderCtx;
deUint32 m_format;
deUint32 m_dataType;
deUint32 m_hint;
int m_width;
int m_height;
glu::Texture2D* m_texture;
TextureRenderer m_renderer;
};
Texture2DGenMipmapCase::Texture2DGenMipmapCase (tcu::TestContext& testCtx, glu::RenderContext& renderCtx, const char* name, const char* desc, deUint32 format, deUint32 dataType, deUint32 hint, int width, int height)
: TestCase (testCtx, name, desc)
, m_renderCtx (renderCtx)
, m_format (format)
, m_dataType (dataType)
, m_hint (hint)
, m_width (width)
, m_height (height)
, m_texture (DE_NULL)
, m_renderer (renderCtx, testCtx.getLog(), glu::GLSL_VERSION_300_ES, glu::PRECISION_HIGHP)
{
}
Texture2DGenMipmapCase::~Texture2DGenMipmapCase (void)
{
deinit();
}
void Texture2DGenMipmapCase::init (void)
{
DE_ASSERT(!m_texture);
m_texture = new glu::Texture2D(m_renderCtx, m_format, m_dataType, m_width, m_height);
}
void Texture2DGenMipmapCase::deinit (void)
{
delete m_texture;
m_texture = DE_NULL;
m_renderer.clear();
}
Texture2DGenMipmapCase::IterateResult Texture2DGenMipmapCase::iterate (void)
{
const glw::Functions& gl = m_renderCtx.getFunctions();
const deUint32 minFilter = GL_NEAREST_MIPMAP_NEAREST;
const deUint32 magFilter = GL_NEAREST;
const deUint32 wrapS = GL_CLAMP_TO_EDGE;
const deUint32 wrapT = GL_CLAMP_TO_EDGE;
const int numLevels = deLog2Floor32(de::max(m_width, m_height))+1;
tcu::Texture2D resultTexture (tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8), m_texture->getRefTexture().getWidth(), m_texture->getRefTexture().getHeight());
vector<float> texCoord;
// Initialize texture level 0 with colored grid.
m_texture->getRefTexture().allocLevel(0);
tcu::fillWithGrid(m_texture->getRefTexture().getLevel(0), 8, tcu::Vec4(1.0f, 0.5f, 0.0f, 0.5f), tcu::Vec4(0.0f, 0.0f, 1.0f, 1.0f));
// Upload data and setup params.
m_texture->upload();
gl.bindTexture (GL_TEXTURE_2D, m_texture->getGLTexture());
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, wrapS);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, wrapT);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, minFilter);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, magFilter);
GLU_EXPECT_NO_ERROR(gl.getError(), "After texture setup");
// Generate mipmap.
gl.hint(GL_GENERATE_MIPMAP_HINT, m_hint);
gl.generateMipmap(GL_TEXTURE_2D);
GLU_EXPECT_NO_ERROR(gl.getError(), "glGenerateMipmap()");
// Use (0, 0) -> (1, 1) texture coordinates.
computeQuadTexCoord2D(texCoord, Vec2(0.0f, 0.0f), Vec2(1.0f, 1.0f));
// Fetch resulting texture by rendering.
for (int levelNdx = 0; levelNdx < numLevels; levelNdx++)
{
const int levelWidth = de::max(1, m_width >> levelNdx);
const int levelHeight = de::max(1, m_height >> levelNdx);
const RandomViewport viewport (m_renderCtx.getRenderTarget(), levelWidth, levelHeight, deStringHash(getName()) + levelNdx);
gl.viewport(viewport.x, viewport.y, viewport.width, viewport.height);
m_renderer.renderQuad(0, &texCoord[0], TEXTURETYPE_2D);
resultTexture.allocLevel(levelNdx);
glu::readPixels(m_renderCtx, viewport.x, viewport.y, resultTexture.getLevel(levelNdx));
}
// Compare results
{
const IVec4 framebufferBits = max(getBitsVec(m_renderCtx.getRenderTarget().getPixelFormat())-2, IVec4(0));
const IVec4 formatBits = tcu::getTextureFormatBitDepth(glu::mapGLTransferFormat(m_format, m_dataType));
const tcu::BVec4 formatMask = greaterThan(formatBits, IVec4(0));
const IVec4 cmpBits = select(min(framebufferBits, formatBits), framebufferBits, formatMask);
GenMipmapPrecision comparePrec;
comparePrec.colorMask = getCompareMask(m_renderCtx.getRenderTarget().getPixelFormat());
comparePrec.colorThreshold = tcu::computeFixedPointThreshold(cmpBits);
comparePrec.filterBits = tcu::IVec3(4, 4, 0);
const qpTestResult compareResult = compareGenMipmapResult(m_testCtx.getLog(), resultTexture, m_texture->getRefTexture(), comparePrec);
m_testCtx.setTestResult(compareResult, compareResult == QP_TEST_RESULT_PASS ? "Pass" :
compareResult == QP_TEST_RESULT_QUALITY_WARNING ? "Low-quality method used" :
compareResult == QP_TEST_RESULT_FAIL ? "Image comparison failed" : "");
}
return STOP;
}
// TextureCubeGenMipmapCase
class TextureCubeGenMipmapCase : public tcu::TestCase
{
public:
TextureCubeGenMipmapCase (tcu::TestContext& testCtx, glu::RenderContext& renderCtx, const char* name, const char* desc, deUint32 format, deUint32 dataType, deUint32 hint, int size);
~TextureCubeGenMipmapCase (void);
void init (void);
void deinit (void);
IterateResult iterate (void);
private:
TextureCubeGenMipmapCase (const TextureCubeGenMipmapCase& other);
TextureCubeGenMipmapCase& operator= (const TextureCubeGenMipmapCase& other);
glu::RenderContext& m_renderCtx;
deUint32 m_format;
deUint32 m_dataType;
deUint32 m_hint;
int m_size;
glu::TextureCube* m_texture;
TextureRenderer m_renderer;
};
TextureCubeGenMipmapCase::TextureCubeGenMipmapCase (tcu::TestContext& testCtx, glu::RenderContext& renderCtx, const char* name, const char* desc, deUint32 format, deUint32 dataType, deUint32 hint, int size)
: TestCase (testCtx, name, desc)
, m_renderCtx (renderCtx)
, m_format (format)
, m_dataType (dataType)
, m_hint (hint)
, m_size (size)
, m_texture (DE_NULL)
, m_renderer (renderCtx, testCtx.getLog(), glu::GLSL_VERSION_300_ES, glu::PRECISION_HIGHP)
{
}
TextureCubeGenMipmapCase::~TextureCubeGenMipmapCase (void)
{
deinit();
}
void TextureCubeGenMipmapCase::init (void)
{
if (m_renderCtx.getRenderTarget().getWidth() < 3*m_size || m_renderCtx.getRenderTarget().getHeight() < 2*m_size)
throw tcu::NotSupportedError("Render target size must be at least (" + de::toString(3*m_size) + ", " + de::toString(2*m_size) + ")");
DE_ASSERT(!m_texture);
m_texture = new glu::TextureCube(m_renderCtx, m_format, m_dataType, m_size);
}
void TextureCubeGenMipmapCase::deinit (void)
{
delete m_texture;
m_texture = DE_NULL;
m_renderer.clear();
}
TextureCubeGenMipmapCase::IterateResult TextureCubeGenMipmapCase::iterate (void)
{
const glw::Functions& gl = m_renderCtx.getFunctions();
const deUint32 minFilter = GL_NEAREST_MIPMAP_NEAREST;
const deUint32 magFilter = GL_NEAREST;
const deUint32 wrapS = GL_CLAMP_TO_EDGE;
const deUint32 wrapT = GL_CLAMP_TO_EDGE;
tcu::TextureCube resultTexture (tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8), m_size);
const int numLevels = deLog2Floor32(m_size)+1;
vector<float> texCoord;
// Initialize texture level 0 with colored grid.
for (int face = 0; face < tcu::CUBEFACE_LAST; face++)
{
Vec4 ca, cb; // Grid colors.
switch (face)
{
case 0: ca = Vec4(1.0f, 0.3f, 0.0f, 0.7f); cb = Vec4(0.0f, 0.0f, 1.0f, 1.0f); break;
case 1: ca = Vec4(0.0f, 1.0f, 0.5f, 0.5f); cb = Vec4(1.0f, 0.0f, 0.0f, 1.0f); break;
case 2: ca = Vec4(0.7f, 0.0f, 1.0f, 0.3f); cb = Vec4(0.0f, 1.0f, 0.0f, 1.0f); break;
case 3: ca = Vec4(0.0f, 0.3f, 1.0f, 1.0f); cb = Vec4(1.0f, 0.0f, 0.0f, 0.7f); break;
case 4: ca = Vec4(1.0f, 0.0f, 0.5f, 1.0f); cb = Vec4(0.0f, 1.0f, 0.0f, 0.5f); break;
case 5: ca = Vec4(0.7f, 1.0f, 0.0f, 1.0f); cb = Vec4(0.0f, 0.0f, 1.0f, 0.3f); break;
}
m_texture->getRefTexture().allocLevel((tcu::CubeFace)face, 0);
fillWithGrid(m_texture->getRefTexture().getLevelFace(0, (tcu::CubeFace)face), 8, ca, cb);
}
// Upload data and setup params.
m_texture->upload();
gl.bindTexture (GL_TEXTURE_CUBE_MAP, m_texture->getGLTexture());
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_S, wrapS);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_T, wrapT);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MIN_FILTER, minFilter);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAG_FILTER, magFilter);
GLU_EXPECT_NO_ERROR(gl.getError(), "After texture setup");
// Generate mipmap.
gl.hint(GL_GENERATE_MIPMAP_HINT, m_hint);
gl.generateMipmap(GL_TEXTURE_CUBE_MAP);
GLU_EXPECT_NO_ERROR(gl.getError(), "glGenerateMipmap()");
// Render all levels.
for (int levelNdx = 0; levelNdx < numLevels; levelNdx++)
{
const int levelWidth = de::max(1, m_size >> levelNdx);
const int levelHeight = de::max(1, m_size >> levelNdx);
for (int faceNdx = 0; faceNdx < tcu::CUBEFACE_LAST; faceNdx++)
{
const RandomViewport viewport (m_renderCtx.getRenderTarget(), levelWidth*3, levelHeight*2, deStringHash(getName()) ^ deInt32Hash(levelNdx + faceNdx));
const tcu::CubeFace face = tcu::CubeFace(faceNdx);
computeQuadTexCoordCube(texCoord, face);
gl.viewport(viewport.x, viewport.y, levelWidth, levelHeight);
m_renderer.renderQuad(0, &texCoord[0], TEXTURETYPE_CUBE);
resultTexture.allocLevel(face, levelNdx);
glu::readPixels(m_renderCtx, viewport.x, viewport.y, resultTexture.getLevelFace(levelNdx, face));
}
}
// Compare results
{
const IVec4 framebufferBits = max(getBitsVec(m_renderCtx.getRenderTarget().getPixelFormat())-2, IVec4(0));
const IVec4 formatBits = tcu::getTextureFormatBitDepth(glu::mapGLTransferFormat(m_format, m_dataType));
const tcu::BVec4 formatMask = greaterThan(formatBits, IVec4(0));
const IVec4 cmpBits = select(min(framebufferBits, formatBits), framebufferBits, formatMask);
GenMipmapPrecision comparePrec;
comparePrec.colorMask = getCompareMask(m_renderCtx.getRenderTarget().getPixelFormat());
comparePrec.colorThreshold = tcu::computeFixedPointThreshold(cmpBits);
comparePrec.filterBits = tcu::IVec3(4, 4, 0);
const qpTestResult compareResult = compareGenMipmapResult(m_testCtx.getLog(), resultTexture, m_texture->getRefTexture(), comparePrec);
m_testCtx.setTestResult(compareResult, compareResult == QP_TEST_RESULT_PASS ? "Pass" :
compareResult == QP_TEST_RESULT_QUALITY_WARNING ? "Low-quality method used" :
compareResult == QP_TEST_RESULT_FAIL ? "Image comparison failed" : "");
}
return STOP;
}
// Texture3DMipmapCase
class Texture3DMipmapCase : public TestCase
{
public:
Texture3DMipmapCase (Context& context,
const char* name,
const char* desc,
CoordType coordType,
deUint32 minFilter,
deUint32 wrapS,
deUint32 wrapT,
deUint32 wrapR,
deUint32 format,
int width,
int height,
int depth);
~Texture3DMipmapCase (void);
void init (void);
void deinit (void);
IterateResult iterate (void);
private:
Texture3DMipmapCase (const Texture3DMipmapCase& other);
Texture3DMipmapCase& operator= (const Texture3DMipmapCase& other);
CoordType m_coordType;
deUint32 m_minFilter;
deUint32 m_wrapS;
deUint32 m_wrapT;
deUint32 m_wrapR;
deUint32 m_internalFormat;
int m_width;
int m_height;
int m_depth;
glu::Texture3D* m_texture;
TextureTestUtil::TextureRenderer m_renderer;
};
Texture3DMipmapCase::Texture3DMipmapCase (Context& context, const char* name, const char* desc, CoordType coordType, deUint32 minFilter, deUint32 wrapS, deUint32 wrapT, deUint32 wrapR, deUint32 format, int width, int height, int depth)
: TestCase (context, name, desc)
, m_coordType (coordType)
, m_minFilter (minFilter)
, m_wrapS (wrapS)
, m_wrapT (wrapT)
, m_wrapR (wrapR)
, m_internalFormat (format)
, m_width (width)
, m_height (height)
, m_depth (depth)
, m_texture (DE_NULL)
, m_renderer (context.getRenderContext(), context.getTestContext().getLog(), glu::GLSL_VERSION_300_ES, glu::PRECISION_HIGHP)
{
}
Texture3DMipmapCase::~Texture3DMipmapCase (void)
{
Texture3DMipmapCase::deinit();
}
void Texture3DMipmapCase::init (void)
{
const tcu::TextureFormat& texFmt = glu::mapGLInternalFormat(m_internalFormat);
tcu::TextureFormatInfo fmtInfo = tcu::getTextureFormatInfo(texFmt);
const tcu::Vec4& cScale = fmtInfo.lookupScale;
const tcu::Vec4& cBias = fmtInfo.lookupBias;
int numLevels = deLog2Floor32(de::max(de::max(m_width, m_height), m_depth))+1;
if (m_coordType == COORDTYPE_PROJECTED && m_context.getRenderTarget().getNumSamples() > 0)
throw tcu::NotSupportedError("Projected lookup validation not supported in multisample config");
m_texture = new glu::Texture3D(m_context.getRenderContext(), m_internalFormat, m_width, m_height, m_depth);
// Fill texture with colored grid.
for (int levelNdx = 0; levelNdx < numLevels; levelNdx++)
{
deUint32 step = 0xff / (numLevels-1);
deUint32 inc = deClamp32(step*levelNdx, 0x00, 0xff);
deUint32 dec = 0xff - inc;
deUint32 rgb = (0xff << 16) | (dec << 8) | inc;
deUint32 color = 0xff000000 | rgb;
m_texture->getRefTexture().allocLevel(levelNdx);
tcu::clear(m_texture->getRefTexture().getLevel(levelNdx), tcu::RGBA(color).toVec()*cScale + cBias);
}
m_texture->upload();
}
void Texture3DMipmapCase::deinit (void)
{
delete m_texture;
m_texture = DE_NULL;
m_renderer.clear();
}
static void getBasicTexCoord3D (std::vector<float>& dst, int cellNdx)
{
static const struct
{
float sScale;
float sBias;
float tScale;
float tBias;
float rScale;
float rBias;
} s_params[] =
{
// sScale sBias tScale tBias rScale rBias
{ 0.9f, -0.1f, 0.7f, 0.3f, 0.8f, 0.9f },
{ 1.2f, -0.1f, 1.1f, 0.3f, 1.0f, 0.9f },
{ 1.5f, 0.7f, 0.9f, -0.3f, 1.1f, 0.1f },
{ 1.2f, 0.7f, -2.3f, -0.3f, 1.1f, 0.2f },
{ 1.1f, 0.8f, -1.3f, -0.3f, 2.9f, 0.9f },
{ 3.4f, 0.8f, 4.0f, 0.0f, -3.3f, -1.0f },
{ -3.4f, -0.1f, -4.0f, 0.0f, -5.1f, 1.0f },
{ -4.0f, -0.1f, 3.4f, 0.1f, 5.7f, 0.0f },
{ -5.6f, 0.0f, 0.5f, 1.2f, 3.9f, 4.0f },
{ 5.0f, -2.0f, 3.1f, 1.2f, 5.1f, 0.2f },
{ 2.5f, -2.0f, 6.3f, 3.0f, 5.1f, 0.2f },
{ -8.3f, 0.0f, 7.1f, 3.0f, 2.0f, 0.2f },
{ 3.8f, 0.0f, 9.7f, 1.0f, 7.0f, 0.7f },
{ 13.3f, 0.0f, 7.1f, 3.0f, 2.0f, 0.2f },
{ 16.0f, 8.0f, 12.7f, 1.0f, 17.1f, 0.7f },
{ 15.3f, 0.0f, 20.1f, 3.0f, 33.0f, 3.2f }
};
float sScale = s_params[cellNdx%DE_LENGTH_OF_ARRAY(s_params)].sScale;
float sBias = s_params[cellNdx%DE_LENGTH_OF_ARRAY(s_params)].sBias;
float tScale = s_params[cellNdx%DE_LENGTH_OF_ARRAY(s_params)].tScale;
float tBias = s_params[cellNdx%DE_LENGTH_OF_ARRAY(s_params)].tBias;
float rScale = s_params[cellNdx%DE_LENGTH_OF_ARRAY(s_params)].rScale;
float rBias = s_params[cellNdx%DE_LENGTH_OF_ARRAY(s_params)].rBias;
dst.resize(3*4);
dst[0] = sBias; dst[ 1] = tBias; dst[ 2] = rBias;
dst[3] = sBias; dst[ 4] = tBias+tScale; dst[ 5] = rBias+rScale*0.5f;
dst[6] = sBias+sScale; dst[ 7] = tBias; dst[ 8] = rBias+rScale*0.5f;
dst[9] = sBias+sScale; dst[10] = tBias+tScale; dst[11] = rBias+rScale;
}
static void getAffineTexCoord3D (std::vector<float>& dst, int cellNdx)
{
// Use basic coords as base.
getBasicTexCoord3D(dst, cellNdx);
// Rotate based on cell index.
float angleX = 0.0f + 2.0f*DE_PI * ((float)cellNdx / 16.0f);
float angleY = 1.0f + 2.0f*DE_PI * ((float)cellNdx / 32.0f);
tcu::Mat3 rotMatrix = tcu::rotationMatrixX(angleX) * tcu::rotationMatrixY(angleY);
Vec3 p0 = rotMatrix * Vec3(dst[0], dst[ 1], dst[ 2]);
Vec3 p1 = rotMatrix * Vec3(dst[3], dst[ 4], dst[ 5]);
Vec3 p2 = rotMatrix * Vec3(dst[6], dst[ 7], dst[ 8]);
Vec3 p3 = rotMatrix * Vec3(dst[9], dst[10], dst[11]);
dst[0] = p0.x(); dst[ 1] = p0.y(); dst[ 2] = p0.z();
dst[3] = p1.x(); dst[ 4] = p1.y(); dst[ 5] = p1.z();
dst[6] = p2.x(); dst[ 7] = p2.y(); dst[ 8] = p2.z();
dst[9] = p3.x(); dst[10] = p3.y(); dst[11] = p3.z();
}
Texture3DMipmapCase::IterateResult Texture3DMipmapCase::iterate (void)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
const tcu::Texture3D& refTexture = m_texture->getRefTexture();
const tcu::TextureFormat& texFmt = refTexture.getFormat();
const tcu::TextureFormatInfo fmtInfo = tcu::getTextureFormatInfo(texFmt);
const int texWidth = refTexture.getWidth();
const int texHeight = refTexture.getHeight();
const deUint32 magFilter = GL_NEAREST;
const tcu::RenderTarget& renderTarget = m_context.getRenderContext().getRenderTarget();
const RandomViewport viewport (renderTarget, texWidth*4, texHeight*4, deStringHash(getName()));
const bool isProjected = m_coordType == COORDTYPE_PROJECTED;
const bool useLodBias = m_coordType == COORDTYPE_BASIC_BIAS;
// Viewport is divided into 4x4 grid.
const int gridWidth = 4;
const int gridHeight = 4;
const int cellWidth = viewport.width / gridWidth;
const int cellHeight = viewport.height / gridHeight;
ReferenceParams sampleParams (TEXTURETYPE_3D);
tcu::Surface renderedFrame (viewport.width, viewport.height);
vector<float> texCoord;
// Sampling parameters.
sampleParams.sampler = glu::mapGLSampler(m_wrapS, m_wrapT, m_wrapR, m_minFilter, magFilter);
sampleParams.samplerType = gls::TextureTestUtil::getSamplerType(texFmt);
sampleParams.colorBias = fmtInfo.lookupBias;
sampleParams.colorScale = fmtInfo.lookupScale;
sampleParams.flags = (isProjected ? ReferenceParams::PROJECTED : 0) | (useLodBias ? ReferenceParams::USE_BIAS : 0);
// Bind texture and setup sampler parameters.
gl.bindTexture (GL_TEXTURE_3D, m_texture->getGLTexture());
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_WRAP_S, m_wrapS);
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_WRAP_T, m_wrapT);
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_WRAP_R, m_wrapR);
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_MIN_FILTER, m_minFilter);
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_MAG_FILTER, magFilter);
GLU_EXPECT_NO_ERROR(gl.getError(), "After texture setup");
// Bias values.
static const float s_bias[] = { 1.0f, -2.0f, 0.8f, -0.5f, 1.5f, 0.9f, 2.0f, 4.0f };
// Projection values.
static const Vec4 s_projections[] =
{
Vec4(1.2f, 1.0f, 0.7f, 1.0f),
Vec4(1.3f, 0.8f, 0.6f, 2.0f),
Vec4(0.8f, 1.0f, 1.7f, 0.6f),
Vec4(1.2f, 1.0f, 1.7f, 1.5f)
};
// Render cells.
for (int gridY = 0; gridY < gridHeight; gridY++)
{
for (int gridX = 0; gridX < gridWidth; gridX++)
{
const int curX = cellWidth*gridX;
const int curY = cellHeight*gridY;
const int curW = gridX+1 == gridWidth ? (viewport.width-curX) : cellWidth;
const int curH = gridY+1 == gridHeight ? (viewport.height-curY) : cellHeight;
const int cellNdx = gridY*gridWidth + gridX;
// Compute texcoord.
switch (m_coordType)
{
case COORDTYPE_BASIC_BIAS: // Fall-through.
case COORDTYPE_PROJECTED:
case COORDTYPE_BASIC: getBasicTexCoord3D (texCoord, cellNdx); break;
case COORDTYPE_AFFINE: getAffineTexCoord3D (texCoord, cellNdx); break;
default: DE_ASSERT(DE_FALSE);
}
// Set projection.
if (isProjected)
sampleParams.w = s_projections[cellNdx % DE_LENGTH_OF_ARRAY(s_projections)];
// Set LOD bias.
if (useLodBias)
sampleParams.bias = s_bias[cellNdx % DE_LENGTH_OF_ARRAY(s_bias)];
// Render with GL.
gl.viewport(viewport.x+curX, viewport.y+curY, curW, curH);
m_renderer.renderQuad(0, &texCoord[0], sampleParams);
}
}
// Read result.
glu::readPixels(m_context.getRenderContext(), viewport.x, viewport.y, renderedFrame.getAccess());
// Compare and log
{
const tcu::PixelFormat& pixelFormat = m_context.getRenderTarget().getPixelFormat();
const bool isTrilinear = m_minFilter == GL_NEAREST_MIPMAP_LINEAR || m_minFilter == GL_LINEAR_MIPMAP_LINEAR;
tcu::Surface referenceFrame (viewport.width, viewport.height);
tcu::Surface errorMask (viewport.width, viewport.height);
tcu::LookupPrecision lookupPrec;
tcu::LodPrecision lodPrec;
int numFailedPixels = 0;
lookupPrec.coordBits = tcu::IVec3(20, 20, 20);
lookupPrec.uvwBits = tcu::IVec3(16, 16, 16); // Doesn't really matter since pixels are unicolored.
lookupPrec.colorThreshold = tcu::computeFixedPointThreshold(max(getBitsVec(pixelFormat) - (isTrilinear ? 2 : 1), tcu::IVec4(0)));
lookupPrec.colorMask = getCompareMask(pixelFormat);
lodPrec.derivateBits = 10;
lodPrec.lodBits = isProjected ? 6 : 8;
for (int gridY = 0; gridY < gridHeight; gridY++)
{
for (int gridX = 0; gridX < gridWidth; gridX++)
{
const int curX = cellWidth*gridX;
const int curY = cellHeight*gridY;
const int curW = gridX+1 == gridWidth ? (viewport.width-curX) : cellWidth;
const int curH = gridY+1 == gridHeight ? (viewport.height-curY) : cellHeight;
const int cellNdx = gridY*gridWidth + gridX;
switch (m_coordType)
{
case COORDTYPE_BASIC_BIAS: // Fall-through.
case COORDTYPE_PROJECTED:
case COORDTYPE_BASIC: getBasicTexCoord3D (texCoord, cellNdx); break;
case COORDTYPE_AFFINE: getAffineTexCoord3D (texCoord, cellNdx); break;
default: DE_ASSERT(DE_FALSE);
}
if (isProjected)
sampleParams.w = s_projections[cellNdx % DE_LENGTH_OF_ARRAY(s_projections)];
if (useLodBias)
sampleParams.bias = s_bias[cellNdx % DE_LENGTH_OF_ARRAY(s_bias)];
// Render ideal result
sampleTexture(SurfaceAccess(referenceFrame, pixelFormat, curX, curY, curW, curH),
refTexture, &texCoord[0], sampleParams);
// Compare this cell
numFailedPixels += computeTextureLookupDiff(tcu::getSubregion(renderedFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(referenceFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(errorMask.getAccess(), curX, curY, curW, curH),
m_texture->getRefTexture(), &texCoord[0], sampleParams,
lookupPrec, lodPrec, m_testCtx.getWatchDog());
}
}
if (numFailedPixels > 0)
m_testCtx.getLog() << TestLog::Message << "ERROR: Image verification failed, found " << numFailedPixels << " invalid pixels!" << TestLog::EndMessage;
m_testCtx.getLog() << TestLog::ImageSet("Result", "Verification result")
<< TestLog::Image("Rendered", "Rendered image", renderedFrame);
if (numFailedPixels > 0)
{
m_testCtx.getLog() << TestLog::Image("Reference", "Ideal reference", referenceFrame)
<< TestLog::Image("ErrorMask", "Error mask", errorMask);
}
m_testCtx.getLog() << TestLog::EndImageSet;
{
const bool isOk = numFailedPixels == 0;
m_testCtx.setTestResult(isOk ? QP_TEST_RESULT_PASS : QP_TEST_RESULT_FAIL,
isOk ? "Pass" : "Image verification failed");
}
}
return STOP;
}
// Texture2DLodControlCase + test cases
class Texture2DLodControlCase : public TestCase
{
public:
Texture2DLodControlCase (Context& context, const char* name, const char* desc, deUint32 minFilter);
~Texture2DLodControlCase (void);
void init (void);
void deinit (void);
IterateResult iterate (void);
protected:
virtual void setTextureParams (int cellNdx) = DE_NULL;
virtual void getReferenceParams (ReferenceParams& params, int cellNdx) = DE_NULL;
const int m_texWidth;
const int m_texHeight;
private:
Texture2DLodControlCase (const Texture2DLodControlCase& other);
Texture2DLodControlCase& operator= (const Texture2DLodControlCase& other);
deUint32 m_minFilter;
glu::Texture2D* m_texture;
TextureTestUtil::TextureRenderer m_renderer;
};
Texture2DLodControlCase::Texture2DLodControlCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: TestCase (context, name, desc)
, m_texWidth (64)
, m_texHeight (64)
, m_minFilter (minFilter)
, m_texture (DE_NULL)
, m_renderer (context.getRenderContext(), context.getTestContext().getLog(), glu::GLSL_VERSION_300_ES, glu::PRECISION_HIGHP)
{
}
Texture2DLodControlCase::~Texture2DLodControlCase (void)
{
Texture2DLodControlCase::deinit();
}
void Texture2DLodControlCase::init (void)
{
const deUint32 format = GL_RGBA8;
int numLevels = deLog2Floor32(de::max(m_texWidth, m_texHeight))+1;
m_texture = new glu::Texture2D(m_context.getRenderContext(), format, m_texWidth, m_texHeight);
// Fill texture with colored grid.
for (int levelNdx = 0; levelNdx < numLevels; levelNdx++)
{
deUint32 step = 0xff / (numLevels-1);
deUint32 inc = deClamp32(step*levelNdx, 0x00, 0xff);
deUint32 dec = 0xff - inc;
deUint32 rgb = (inc << 16) | (dec << 8) | 0xff;
deUint32 color = 0xff000000 | rgb;
m_texture->getRefTexture().allocLevel(levelNdx);
tcu::clear(m_texture->getRefTexture().getLevel(levelNdx), tcu::RGBA(color).toVec());
}
}
void Texture2DLodControlCase::deinit (void)
{
delete m_texture;
m_texture = DE_NULL;
m_renderer.clear();
}
Texture2DLodControlCase::IterateResult Texture2DLodControlCase::iterate (void)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
const deUint32 wrapS = GL_REPEAT;
const deUint32 wrapT = GL_REPEAT;
const deUint32 magFilter = GL_NEAREST;
const tcu::Texture2D& refTexture = m_texture->getRefTexture();
const int texWidth = refTexture.getWidth();
const int texHeight = refTexture.getHeight();
const tcu::RenderTarget& renderTarget = m_context.getRenderContext().getRenderTarget();
const RandomViewport viewport (renderTarget, texWidth*4, texHeight*4, deStringHash(getName()));
ReferenceParams sampleParams (gls::TextureTestUtil::TEXTURETYPE_2D, glu::mapGLSampler(wrapS, wrapT, m_minFilter, magFilter));
vector<float> texCoord;
tcu::Surface renderedFrame (viewport.width, viewport.height);
// Viewport is divided into 4x4 grid.
const int gridWidth = 4;
const int gridHeight = 4;
const int cellWidth = viewport.width / gridWidth;
const int cellHeight = viewport.height / gridHeight;
// Upload texture data.
m_texture->upload();
// Bind gradient texture and setup sampler parameters.
gl.bindTexture (GL_TEXTURE_2D, m_texture->getGLTexture());
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, wrapS);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, wrapT);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, m_minFilter);
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, magFilter);
GLU_EXPECT_NO_ERROR(gl.getError(), "After texture setup");
// Render cells.
for (int gridY = 0; gridY < gridHeight; gridY++)
{
for (int gridX = 0; gridX < gridWidth; gridX++)
{
int curX = cellWidth*gridX;
int curY = cellHeight*gridY;
int curW = gridX+1 == gridWidth ? (viewport.width-curX) : cellWidth;
int curH = gridY+1 == gridHeight ? (viewport.height-curY) : cellHeight;
int cellNdx = gridY*gridWidth + gridX;
// Compute texcoord.
getBasicTexCoord2D(texCoord, cellNdx);
// Render with GL.
setTextureParams(cellNdx);
gl.viewport(viewport.x+curX, viewport.y+curY, curW, curH);
m_renderer.renderQuad(0, &texCoord[0], sampleParams);
}
}
glu::readPixels(m_context.getRenderContext(), viewport.x, viewport.y, renderedFrame.getAccess());
GLU_EXPECT_NO_ERROR(gl.getError(), "Read pixels");
// Compare and log.
{
const tcu::PixelFormat& pixelFormat = m_context.getRenderTarget().getPixelFormat();
const bool isTrilinear = m_minFilter == GL_NEAREST_MIPMAP_LINEAR || m_minFilter == GL_LINEAR_MIPMAP_LINEAR;
tcu::Surface referenceFrame (viewport.width, viewport.height);
tcu::Surface errorMask (viewport.width, viewport.height);
tcu::LookupPrecision lookupPrec;
tcu::LodPrecision lodPrec;
int numFailedPixels = 0;
lookupPrec.coordBits = tcu::IVec3(20, 20, 0);
lookupPrec.uvwBits = tcu::IVec3(16, 16, 0); // Doesn't really matter since pixels are unicolored.
lookupPrec.colorThreshold = tcu::computeFixedPointThreshold(max(getBitsVec(pixelFormat) - (isTrilinear ? 2 : 1), tcu::IVec4(0)));
lookupPrec.colorMask = getCompareMask(pixelFormat);
lodPrec.derivateBits = 10;
lodPrec.lodBits = 8;
for (int gridY = 0; gridY < gridHeight; gridY++)
{
for (int gridX = 0; gridX < gridWidth; gridX++)
{
const int curX = cellWidth*gridX;
const int curY = cellHeight*gridY;
const int curW = gridX+1 == gridWidth ? (viewport.width-curX) : cellWidth;
const int curH = gridY+1 == gridHeight ? (viewport.height-curY) : cellHeight;
const int cellNdx = gridY*gridWidth + gridX;
getBasicTexCoord2D(texCoord, cellNdx);
getReferenceParams(sampleParams, cellNdx);
// Render ideal result
sampleTexture(SurfaceAccess(referenceFrame, pixelFormat, curX, curY, curW, curH),
refTexture, &texCoord[0], sampleParams);
// Compare this cell
numFailedPixels += computeTextureLookupDiff(tcu::getSubregion(renderedFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(referenceFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(errorMask.getAccess(), curX, curY, curW, curH),
m_texture->getRefTexture(), &texCoord[0], sampleParams,
lookupPrec, lodPrec, m_testCtx.getWatchDog());
}
}
if (numFailedPixels > 0)
m_testCtx.getLog() << TestLog::Message << "ERROR: Image verification failed, found " << numFailedPixels << " invalid pixels!" << TestLog::EndMessage;
m_testCtx.getLog() << TestLog::ImageSet("Result", "Verification result")
<< TestLog::Image("Rendered", "Rendered image", renderedFrame);
if (numFailedPixels > 0)
{
m_testCtx.getLog() << TestLog::Image("Reference", "Ideal reference", referenceFrame)
<< TestLog::Image("ErrorMask", "Error mask", errorMask);
}
m_testCtx.getLog() << TestLog::EndImageSet;
{
const bool isOk = numFailedPixels == 0;
m_testCtx.setTestResult(isOk ? QP_TEST_RESULT_PASS : QP_TEST_RESULT_FAIL,
isOk ? "Pass" : "Image verification failed");
}
}
return STOP;
}
class Texture2DMinLodCase : public Texture2DLodControlCase
{
public:
Texture2DMinLodCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: Texture2DLodControlCase(context, name, desc, minFilter)
{
}
protected:
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_LOD, getMinLodForCell(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.minLod = getMinLodForCell(cellNdx);
}
};
class Texture2DMaxLodCase : public Texture2DLodControlCase
{
public:
Texture2DMaxLodCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: Texture2DLodControlCase(context, name, desc, minFilter)
{
}
protected:
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAX_LOD, getMaxLodForCell(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.maxLod = getMaxLodForCell(cellNdx);
}
};
class Texture2DBaseLevelCase : public Texture2DLodControlCase
{
public:
Texture2DBaseLevelCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: Texture2DLodControlCase(context, name, desc, minFilter)
{
}
protected:
int getBaseLevel (int cellNdx) const
{
const int numLevels = deLog2Floor32(de::max(m_texWidth, m_texHeight))+1;
const int baseLevel = (deInt32Hash(cellNdx) ^ deStringHash(getName()) ^ 0xac2f274a) % numLevels;
return baseLevel;
}
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_BASE_LEVEL, getBaseLevel(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.baseLevel = getBaseLevel(cellNdx);
}
};
class Texture2DMaxLevelCase : public Texture2DLodControlCase
{
public:
Texture2DMaxLevelCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: Texture2DLodControlCase(context, name, desc, minFilter)
{
}
protected:
int getMaxLevel (int cellNdx) const
{
const int numLevels = deLog2Floor32(de::max(m_texWidth, m_texHeight))+1;
const int maxLevel = (deInt32Hash(cellNdx) ^ deStringHash(getName()) ^ 0x82cfa4e) % numLevels;
return maxLevel;
}
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, getMaxLevel(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.maxLevel = getMaxLevel(cellNdx);
}
};
// TextureCubeLodControlCase + test cases
class TextureCubeLodControlCase : public TestCase
{
public:
TextureCubeLodControlCase (Context& context, const char* name, const char* desc, deUint32 minFilter);
~TextureCubeLodControlCase (void);
void init (void);
void deinit (void);
IterateResult iterate (void);
protected:
virtual void setTextureParams (int cellNdx) = DE_NULL;
virtual void getReferenceParams (ReferenceParams& params, int cellNdx) = DE_NULL;
const int m_texSize;
private:
TextureCubeLodControlCase (const TextureCubeLodControlCase& other);
TextureCubeLodControlCase& operator= (const TextureCubeLodControlCase& other);
deUint32 m_minFilter;
glu::TextureCube* m_texture;
TextureTestUtil::TextureRenderer m_renderer;
};
TextureCubeLodControlCase::TextureCubeLodControlCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: TestCase (context, name, desc)
, m_texSize (64)
, m_minFilter (minFilter)
, m_texture (DE_NULL)
, m_renderer (context.getRenderContext(), context.getTestContext().getLog(), glu::GLSL_VERSION_300_ES, glu::PRECISION_HIGHP)
{
}
TextureCubeLodControlCase::~TextureCubeLodControlCase (void)
{
deinit();
}
void TextureCubeLodControlCase::init (void)
{
const deUint32 format = GL_RGBA8;
const int numLevels = deLog2Floor32(m_texSize)+1;
m_texture = new glu::TextureCube(m_context.getRenderContext(), format, m_texSize);
// Fill texture with colored grid.
for (int faceNdx = 0; faceNdx < tcu::CUBEFACE_LAST; faceNdx++)
{
for (int levelNdx = 0; levelNdx < numLevels; levelNdx++)
{
deUint32 step = 0xff / (numLevels-1);
deUint32 inc = deClamp32(step*levelNdx, 0x00, 0xff);
deUint32 dec = 0xff - inc;
deUint32 rgb = 0;
switch (faceNdx)
{
case 0: rgb = (inc << 16) | (dec << 8) | 255; break;
case 1: rgb = (255 << 16) | (inc << 8) | dec; break;
case 2: rgb = (dec << 16) | (255 << 8) | inc; break;
case 3: rgb = (dec << 16) | (inc << 8) | 255; break;
case 4: rgb = (255 << 16) | (dec << 8) | inc; break;
case 5: rgb = (inc << 16) | (255 << 8) | dec; break;
}
deUint32 color = 0xff000000 | rgb;
m_texture->getRefTexture().allocLevel((tcu::CubeFace)faceNdx, levelNdx);
tcu::clear(m_texture->getRefTexture().getLevelFace(levelNdx, (tcu::CubeFace)faceNdx), tcu::RGBA(color).toVec());
}
}
}
void TextureCubeLodControlCase::deinit (void)
{
delete m_texture;
m_texture = DE_NULL;
m_renderer.clear();
}
TextureCubeLodControlCase::IterateResult TextureCubeLodControlCase::iterate (void)
{
const deUint32 wrapS = GL_CLAMP_TO_EDGE;
const deUint32 wrapT = GL_CLAMP_TO_EDGE;
const deUint32 magFilter = GL_NEAREST;
const int texWidth = m_texture->getRefTexture().getSize();
const int texHeight = m_texture->getRefTexture().getSize();
const int defViewportWidth = texWidth*2;
const int defViewportHeight = texHeight*2;
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
const RandomViewport viewport (m_context.getRenderTarget(), defViewportWidth, defViewportHeight, deStringHash(getName()));
vector<float> texCoord;
tcu::Surface renderedFrame (viewport.width, viewport.height);
// Upload texture data.
m_texture->upload();
// Bind gradient texture and setup sampler parameters.
gl.bindTexture (GL_TEXTURE_CUBE_MAP, m_texture->getGLTexture());
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_S, wrapS);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_T, wrapT);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MIN_FILTER, m_minFilter);
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAG_FILTER, magFilter);
GLU_EXPECT_NO_ERROR(gl.getError(), "After texture setup");
// Compute grid.
vector<tcu::IVec4> gridLayout;
computeGridLayout(gridLayout, viewport.width, viewport.height);
for (int cellNdx = 0; cellNdx < (int)gridLayout.size(); cellNdx++)
{
const int curX = gridLayout[cellNdx].x();
const int curY = gridLayout[cellNdx].y();
const int curW = gridLayout[cellNdx].z();
const int curH = gridLayout[cellNdx].w();
const tcu::CubeFace cubeFace = (tcu::CubeFace)(cellNdx % tcu::CUBEFACE_LAST);
RenderParams params (TEXTURETYPE_CUBE);
TextureTestUtil::computeQuadTexCoordCube(texCoord, cubeFace);
setTextureParams(cellNdx);
// Render with GL.
gl.viewport(viewport.x+curX, viewport.y+curY, curW, curH);
m_renderer.renderQuad(0, &texCoord[0], params);
GLU_EXPECT_NO_ERROR(gl.getError(), "Draw");
}
// Read result.
glu::readPixels(m_context.getRenderContext(), viewport.x, viewport.y, renderedFrame.getAccess());
GLU_EXPECT_NO_ERROR(gl.getError(), "Read pixels");
// Render reference and compare
{
tcu::Surface referenceFrame (viewport.width, viewport.height);
tcu::Surface errorMask (viewport.width, viewport.height);
int numFailedPixels = 0;
ReferenceParams params (TEXTURETYPE_CUBE);
tcu::LookupPrecision lookupPrec;
tcu::LodPrecision lodPrec;
// Params for rendering reference
params.sampler = glu::mapGLSampler(wrapS, wrapT, m_minFilter, magFilter);
params.sampler.seamlessCubeMap = true;
params.lodMode = LODMODE_EXACT;
// Comparison parameters
lookupPrec.colorMask = getCompareMask(m_context.getRenderTarget().getPixelFormat());
lookupPrec.colorThreshold = tcu::computeFixedPointThreshold(max(getBitsVec(m_context.getRenderTarget().getPixelFormat())-2, IVec4(0)));
lookupPrec.coordBits = tcu::IVec3(10);
lookupPrec.uvwBits = tcu::IVec3(5,5,0);
lodPrec.derivateBits = 10;
lodPrec.lodBits = 6;
for (int cellNdx = 0; cellNdx < (int)gridLayout.size(); cellNdx++)
{
const int curX = gridLayout[cellNdx].x();
const int curY = gridLayout[cellNdx].y();
const int curW = gridLayout[cellNdx].z();
const int curH = gridLayout[cellNdx].w();
const tcu::CubeFace cubeFace = (tcu::CubeFace)(cellNdx % tcu::CUBEFACE_LAST);
computeQuadTexCoordCube(texCoord, cubeFace);
getReferenceParams(params, cellNdx);
// Render ideal reference.
{
SurfaceAccess idealDst(referenceFrame, m_context.getRenderTarget().getPixelFormat(), curX, curY, curW, curH);
sampleTexture(idealDst, m_texture->getRefTexture(), &texCoord[0], params);
}
// Compare this cell
numFailedPixels += computeTextureLookupDiff(tcu::getSubregion(renderedFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(referenceFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(errorMask.getAccess(), curX, curY, curW, curH),
m_texture->getRefTexture(), &texCoord[0], params,
lookupPrec, lodPrec, m_testCtx.getWatchDog());
}
if (numFailedPixels > 0)
m_testCtx.getLog() << TestLog::Message << "ERROR: Image verification failed, found " << numFailedPixels << " invalid pixels!" << TestLog::EndMessage;
m_testCtx.getLog() << TestLog::ImageSet("Result", "Verification result")
<< TestLog::Image("Rendered", "Rendered image", renderedFrame);
if (numFailedPixels > 0)
{
m_testCtx.getLog() << TestLog::Image("Reference", "Ideal reference", referenceFrame)
<< TestLog::Image("ErrorMask", "Error mask", errorMask);
}
m_testCtx.getLog() << TestLog::EndImageSet;
{
const bool isOk = numFailedPixels == 0;
m_testCtx.setTestResult(isOk ? QP_TEST_RESULT_PASS : QP_TEST_RESULT_FAIL,
isOk ? "Pass" : "Image verification failed");
}
}
return STOP;
}
class TextureCubeMinLodCase : public TextureCubeLodControlCase
{
public:
TextureCubeMinLodCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: TextureCubeLodControlCase(context, name, desc, minFilter)
{
}
protected:
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MIN_LOD, getMinLodForCell(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.minLod = getMinLodForCell(cellNdx);
}
};
class TextureCubeMaxLodCase : public TextureCubeLodControlCase
{
public:
TextureCubeMaxLodCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: TextureCubeLodControlCase(context, name, desc, minFilter)
{
}
protected:
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAX_LOD, getMaxLodForCell(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.maxLod = getMaxLodForCell(cellNdx);
}
};
class TextureCubeBaseLevelCase : public TextureCubeLodControlCase
{
public:
TextureCubeBaseLevelCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: TextureCubeLodControlCase(context, name, desc, minFilter)
{
}
protected:
int getBaseLevel (int cellNdx) const
{
const int numLevels = deLog2Floor32(m_texSize)+1;
const int baseLevel = (deInt32Hash(cellNdx) ^ deStringHash(getName()) ^ 0x23fae13) % numLevels;
return baseLevel;
}
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_BASE_LEVEL, getBaseLevel(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.baseLevel = getBaseLevel(cellNdx);
}
};
class TextureCubeMaxLevelCase : public TextureCubeLodControlCase
{
public:
TextureCubeMaxLevelCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: TextureCubeLodControlCase(context, name, desc, minFilter)
{
}
protected:
int getMaxLevel (int cellNdx) const
{
const int numLevels = deLog2Floor32(m_texSize)+1;
const int maxLevel = (deInt32Hash(cellNdx) ^ deStringHash(getName()) ^ 0x974e21) % numLevels;
return maxLevel;
}
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAX_LEVEL, getMaxLevel(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.maxLevel = getMaxLevel(cellNdx);
}
};
// Texture3DLodControlCase + test cases
class Texture3DLodControlCase : public TestCase
{
public:
Texture3DLodControlCase (Context& context, const char* name, const char* desc, deUint32 minFilter);
~Texture3DLodControlCase (void);
void init (void);
void deinit (void);
IterateResult iterate (void);
protected:
virtual void setTextureParams (int cellNdx) = DE_NULL;
virtual void getReferenceParams (ReferenceParams& params, int cellNdx) = DE_NULL;
const int m_texWidth;
const int m_texHeight;
const int m_texDepth;
private:
Texture3DLodControlCase (const Texture3DLodControlCase& other);
Texture3DLodControlCase& operator= (const Texture3DLodControlCase& other);
deUint32 m_minFilter;
glu::Texture3D* m_texture;
TextureTestUtil::TextureRenderer m_renderer;
};
Texture3DLodControlCase::Texture3DLodControlCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: TestCase (context, name, desc)
, m_texWidth (32)
, m_texHeight (32)
, m_texDepth (32)
, m_minFilter (minFilter)
, m_texture (DE_NULL)
, m_renderer (context.getRenderContext(), context.getTestContext().getLog(), glu::GLSL_VERSION_300_ES, glu::PRECISION_HIGHP)
{
}
Texture3DLodControlCase::~Texture3DLodControlCase (void)
{
Texture3DLodControlCase::deinit();
}
void Texture3DLodControlCase::init (void)
{
const deUint32 format = GL_RGBA8;
const tcu::TextureFormat& texFmt = glu::mapGLInternalFormat(format);
tcu::TextureFormatInfo fmtInfo = tcu::getTextureFormatInfo(texFmt);
const tcu::Vec4& cScale = fmtInfo.lookupScale;
const tcu::Vec4& cBias = fmtInfo.lookupBias;
int numLevels = deLog2Floor32(de::max(de::max(m_texWidth, m_texHeight), m_texDepth))+1;
m_texture = new glu::Texture3D(m_context.getRenderContext(), format, m_texWidth, m_texHeight, m_texDepth);
// Fill texture with colored grid.
for (int levelNdx = 0; levelNdx < numLevels; levelNdx++)
{
deUint32 step = 0xff / (numLevels-1);
deUint32 inc = deClamp32(step*levelNdx, 0x00, 0xff);
deUint32 dec = 0xff - inc;
deUint32 rgb = (inc << 16) | (dec << 8) | 0xff;
deUint32 color = 0xff000000 | rgb;
m_texture->getRefTexture().allocLevel(levelNdx);
tcu::clear(m_texture->getRefTexture().getLevel(levelNdx), tcu::RGBA(color).toVec()*cScale + cBias);
}
m_texture->upload();
}
void Texture3DLodControlCase::deinit (void)
{
delete m_texture;
m_texture = DE_NULL;
m_renderer.clear();
}
Texture3DLodControlCase::IterateResult Texture3DLodControlCase::iterate (void)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
const deUint32 wrapS = GL_CLAMP_TO_EDGE;
const deUint32 wrapT = GL_CLAMP_TO_EDGE;
const deUint32 wrapR = GL_CLAMP_TO_EDGE;
const deUint32 magFilter = GL_NEAREST;
const tcu::Texture3D& refTexture = m_texture->getRefTexture();
const tcu::TextureFormat& texFmt = refTexture.getFormat();
const tcu::TextureFormatInfo fmtInfo = tcu::getTextureFormatInfo(texFmt);
const int texWidth = refTexture.getWidth();
const int texHeight = refTexture.getHeight();
const tcu::RenderTarget& renderTarget = m_context.getRenderContext().getRenderTarget();
const RandomViewport viewport (renderTarget, texWidth*4, texHeight*4, deStringHash(getName()));
// Viewport is divided into 4x4 grid.
const int gridWidth = 4;
const int gridHeight = 4;
const int cellWidth = viewport.width / gridWidth;
const int cellHeight = viewport.height / gridHeight;
tcu::Surface renderedFrame (viewport.width, viewport.height);
vector<float> texCoord;
ReferenceParams sampleParams (gls::TextureTestUtil::TEXTURETYPE_3D);
// Sampling parameters.
sampleParams.sampler = glu::mapGLSampler(wrapS, wrapT, wrapR, m_minFilter, magFilter);
sampleParams.samplerType = gls::TextureTestUtil::getSamplerType(texFmt);
sampleParams.colorBias = fmtInfo.lookupBias;
sampleParams.colorScale = fmtInfo.lookupScale;
// Bind texture and setup sampler parameters.
gl.bindTexture (GL_TEXTURE_3D, m_texture->getGLTexture());
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_WRAP_S, wrapS);
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_WRAP_T, wrapT);
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_WRAP_R, wrapR);
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_MIN_FILTER, m_minFilter);
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_MAG_FILTER, magFilter);
GLU_EXPECT_NO_ERROR(gl.getError(), "After texture setup");
// Render cells.
for (int gridY = 0; gridY < gridHeight; gridY++)
{
for (int gridX = 0; gridX < gridWidth; gridX++)
{
int curX = cellWidth*gridX;
int curY = cellHeight*gridY;
int curW = gridX+1 == gridWidth ? (viewport.width-curX) : cellWidth;
int curH = gridY+1 == gridHeight ? (viewport.height-curY) : cellHeight;
int cellNdx = gridY*gridWidth + gridX;
// Compute texcoord.
getBasicTexCoord3D(texCoord, cellNdx);
setTextureParams(cellNdx);
// Render with GL.
gl.viewport(viewport.x+curX, viewport.y+curY, curW, curH);
m_renderer.renderQuad(0, &texCoord[0], sampleParams);
}
}
// Read result.
glu::readPixels(m_context.getRenderContext(), viewport.x, viewport.y, renderedFrame.getAccess());
// Compare and log
{
const tcu::PixelFormat& pixelFormat = m_context.getRenderTarget().getPixelFormat();
const bool isTrilinear = m_minFilter == GL_NEAREST_MIPMAP_LINEAR || m_minFilter == GL_LINEAR_MIPMAP_LINEAR;
tcu::Surface referenceFrame (viewport.width, viewport.height);
tcu::Surface errorMask (viewport.width, viewport.height);
tcu::LookupPrecision lookupPrec;
tcu::LodPrecision lodPrec;
int numFailedPixels = 0;
lookupPrec.coordBits = tcu::IVec3(20, 20, 20);
lookupPrec.uvwBits = tcu::IVec3(16, 16, 16); // Doesn't really matter since pixels are unicolored.
lookupPrec.colorThreshold = tcu::computeFixedPointThreshold(max(getBitsVec(pixelFormat) - (isTrilinear ? 2 : 1), tcu::IVec4(0)));
lookupPrec.colorMask = getCompareMask(pixelFormat);
lodPrec.derivateBits = 10;
lodPrec.lodBits = 8;
for (int gridY = 0; gridY < gridHeight; gridY++)
{
for (int gridX = 0; gridX < gridWidth; gridX++)
{
const int curX = cellWidth*gridX;
const int curY = cellHeight*gridY;
const int curW = gridX+1 == gridWidth ? (viewport.width-curX) : cellWidth;
const int curH = gridY+1 == gridHeight ? (viewport.height-curY) : cellHeight;
const int cellNdx = gridY*gridWidth + gridX;
getBasicTexCoord3D(texCoord, cellNdx);
getReferenceParams(sampleParams, cellNdx);
// Render ideal result
sampleTexture(SurfaceAccess(referenceFrame, pixelFormat, curX, curY, curW, curH),
refTexture, &texCoord[0], sampleParams);
// Compare this cell
numFailedPixels += computeTextureLookupDiff(tcu::getSubregion(renderedFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(referenceFrame.getAccess(), curX, curY, curW, curH),
tcu::getSubregion(errorMask.getAccess(), curX, curY, curW, curH),
m_texture->getRefTexture(), &texCoord[0], sampleParams,
lookupPrec, lodPrec, m_testCtx.getWatchDog());
}
}
if (numFailedPixels > 0)
m_testCtx.getLog() << TestLog::Message << "ERROR: Image verification failed, found " << numFailedPixels << " invalid pixels!" << TestLog::EndMessage;
m_testCtx.getLog() << TestLog::ImageSet("Result", "Verification result")
<< TestLog::Image("Rendered", "Rendered image", renderedFrame);
if (numFailedPixels > 0)
{
m_testCtx.getLog() << TestLog::Image("Reference", "Ideal reference", referenceFrame)
<< TestLog::Image("ErrorMask", "Error mask", errorMask);
}
m_testCtx.getLog() << TestLog::EndImageSet;
{
const bool isOk = numFailedPixels == 0;
m_testCtx.setTestResult(isOk ? QP_TEST_RESULT_PASS : QP_TEST_RESULT_FAIL,
isOk ? "Pass" : "Image verification failed");
}
}
return STOP;
}
class Texture3DMinLodCase : public Texture3DLodControlCase
{
public:
Texture3DMinLodCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: Texture3DLodControlCase(context, name, desc, minFilter)
{
}
protected:
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameterf(GL_TEXTURE_3D, GL_TEXTURE_MIN_LOD, getMinLodForCell(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.minLod = getMinLodForCell(cellNdx);
}
};
class Texture3DMaxLodCase : public Texture3DLodControlCase
{
public:
Texture3DMaxLodCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: Texture3DLodControlCase(context, name, desc, minFilter)
{
}
protected:
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameterf(GL_TEXTURE_3D, GL_TEXTURE_MAX_LOD, getMaxLodForCell(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.maxLod = getMaxLodForCell(cellNdx);
}
};
class Texture3DBaseLevelCase : public Texture3DLodControlCase
{
public:
Texture3DBaseLevelCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: Texture3DLodControlCase(context, name, desc, minFilter)
{
}
protected:
int getBaseLevel (int cellNdx) const
{
const int numLevels = deLog2Floor32(de::max(m_texWidth, de::max(m_texHeight, m_texDepth)))+1;
const int baseLevel = (deInt32Hash(cellNdx) ^ deStringHash(getName()) ^ 0x7347e9) % numLevels;
return baseLevel;
}
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_BASE_LEVEL, getBaseLevel(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.baseLevel = getBaseLevel(cellNdx);
}
};
class Texture3DMaxLevelCase : public Texture3DLodControlCase
{
public:
Texture3DMaxLevelCase (Context& context, const char* name, const char* desc, deUint32 minFilter)
: Texture3DLodControlCase(context, name, desc, minFilter)
{
}
protected:
int getMaxLevel (int cellNdx) const
{
const int numLevels = deLog2Floor32(de::max(m_texWidth, de::max(m_texHeight, m_texDepth)))+1;
const int maxLevel = (deInt32Hash(cellNdx) ^ deStringHash(getName()) ^ 0x9111e7) % numLevels;
return maxLevel;
}
void setTextureParams (int cellNdx)
{
const glw::Functions& gl = m_context.getRenderContext().getFunctions();
gl.texParameteri(GL_TEXTURE_3D, GL_TEXTURE_MAX_LEVEL, getMaxLevel(cellNdx));
}
void getReferenceParams (ReferenceParams& params, int cellNdx)
{
params.maxLevel = getMaxLevel(cellNdx);
}
};
TextureMipmapTests::TextureMipmapTests (Context& context)
: TestCaseGroup(context, "mipmap", "Mipmapping tests")
{
}
TextureMipmapTests::~TextureMipmapTests (void)
{
}
void TextureMipmapTests::init (void)
{
tcu::TestCaseGroup* group2D = new tcu::TestCaseGroup(m_testCtx, "2d", "2D Texture Mipmapping");
tcu::TestCaseGroup* groupCube = new tcu::TestCaseGroup(m_testCtx, "cube", "Cube Map Mipmapping");
tcu::TestCaseGroup* group3D = new tcu::TestCaseGroup(m_testCtx, "3d", "3D Texture Mipmapping");
addChild(group2D);
addChild(groupCube);
addChild(group3D);
static const struct
{
const char* name;
deUint32 mode;
} wrapModes[] =
{
{ "clamp", GL_CLAMP_TO_EDGE },
{ "repeat", GL_REPEAT },
{ "mirror", GL_MIRRORED_REPEAT }
};
static const struct
{
const char* name;
deUint32 mode;
} minFilterModes[] =
{
{ "nearest_nearest", GL_NEAREST_MIPMAP_NEAREST },
{ "linear_nearest", GL_LINEAR_MIPMAP_NEAREST },
{ "nearest_linear", GL_NEAREST_MIPMAP_LINEAR },
{ "linear_linear", GL_LINEAR_MIPMAP_LINEAR }
};
static const struct
{
CoordType type;
const char* name;
const char* desc;
} coordTypes[] =
{
{ COORDTYPE_BASIC, "basic", "Mipmapping with translated and scaled coordinates" },
{ COORDTYPE_AFFINE, "affine", "Mipmapping with affine coordinate transform" },
{ COORDTYPE_PROJECTED, "projected", "Mipmapping with perspective projection" }
};
static const struct
{
const char* name;
deUint32 format;
deUint32 dataType;
} formats[] =
{
{ "a8", GL_ALPHA, GL_UNSIGNED_BYTE },
{ "l8", GL_LUMINANCE, GL_UNSIGNED_BYTE },
{ "la88", GL_LUMINANCE_ALPHA, GL_UNSIGNED_BYTE },
{ "rgb565", GL_RGB, GL_UNSIGNED_SHORT_5_6_5 },
{ "rgb888", GL_RGB, GL_UNSIGNED_BYTE },
{ "rgba4444", GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4 },
{ "rgba5551", GL_RGBA, GL_UNSIGNED_SHORT_5_5_5_1 },
{ "rgba8888", GL_RGBA, GL_UNSIGNED_BYTE }
};
static const struct
{
const char* name;
deUint32 hint;
} genHints[] =
{
{ "fastest", GL_FASTEST },
{ "nicest", GL_NICEST }
};
static const struct
{
const char* name;
int width;
int height;
} tex2DSizes[] =
{
{ DE_NULL, 64, 64 }, // Default.
{ "npot", 63, 57 },
{ "non_square", 32, 64 }
};
static const struct
{
const char* name;
int width;
int height;
int depth;
} tex3DSizes[] =
{
{ DE_NULL, 32, 32, 32 }, // Default.
{ "npot", 33, 29, 27 }
};
const int cubeMapSize = 64;
static const struct
{
CoordType type;
const char* name;
const char* desc;
} cubeCoordTypes[] =
{
{ COORDTYPE_BASIC, "basic", "Mipmapping with translated and scaled coordinates" },
{ COORDTYPE_PROJECTED, "projected", "Mipmapping with perspective projection" },
{ COORDTYPE_BASIC_BIAS, "bias", "User-supplied bias value" }
};
// 2D cases.
for (int coordType = 0; coordType < DE_LENGTH_OF_ARRAY(coordTypes); coordType++)
{
tcu::TestCaseGroup* coordTypeGroup = new tcu::TestCaseGroup(m_testCtx, coordTypes[coordType].name, coordTypes[coordType].desc);
group2D->addChild(coordTypeGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
{
for (int wrapMode = 0; wrapMode < DE_LENGTH_OF_ARRAY(wrapModes); wrapMode++)
{
// Add non_square variants to basic cases only.
int sizeEnd = coordTypes[coordType].type == COORDTYPE_BASIC ? DE_LENGTH_OF_ARRAY(tex2DSizes) : 1;
for (int size = 0; size < sizeEnd; size++)
{
std::ostringstream name;
name << minFilterModes[minFilter].name
<< "_" << wrapModes[wrapMode].name;
if (tex2DSizes[size].name)
name << "_" << tex2DSizes[size].name;
coordTypeGroup->addChild(new Texture2DMipmapCase(m_testCtx, m_context.getRenderContext(), m_context.getContextInfo(),
name.str().c_str(), "",
coordTypes[coordType].type,
minFilterModes[minFilter].mode,
wrapModes[wrapMode].mode,
wrapModes[wrapMode].mode,
GL_RGBA, GL_UNSIGNED_BYTE,
tex2DSizes[size].width, tex2DSizes[size].height));
}
}
}
}
// 2D bias variants.
{
tcu::TestCaseGroup* biasGroup = new tcu::TestCaseGroup(m_testCtx, "bias", "User-supplied bias value");
group2D->addChild(biasGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
biasGroup->addChild(new Texture2DMipmapCase(m_testCtx, m_context.getRenderContext(), m_context.getContextInfo(),
minFilterModes[minFilter].name, "",
COORDTYPE_BASIC_BIAS,
minFilterModes[minFilter].mode,
GL_REPEAT, GL_REPEAT,
GL_RGBA, GL_UNSIGNED_BYTE,
tex2DSizes[0].width, tex2DSizes[0].height));
}
// 2D mipmap generation variants.
{
tcu::TestCaseGroup* genMipmapGroup = new tcu::TestCaseGroup(m_testCtx, "generate", "Mipmap generation tests");
group2D->addChild(genMipmapGroup);
for (int format = 0; format < DE_LENGTH_OF_ARRAY(formats); format++)
{
for (int size = 0; size < DE_LENGTH_OF_ARRAY(tex2DSizes); size++)
{
for (int hint = 0; hint < DE_LENGTH_OF_ARRAY(genHints); hint++)
{
std::ostringstream name;
name << formats[format].name;
if (tex2DSizes[size].name)
name << "_" << tex2DSizes[size].name;
name << "_" << genHints[hint].name;
genMipmapGroup->addChild(new Texture2DGenMipmapCase(m_testCtx, m_context.getRenderContext(), name.str().c_str(), "",
formats[format].format, formats[format].dataType, genHints[hint].hint,
tex2DSizes[size].width, tex2DSizes[size].height));
}
}
}
}
// 2D LOD controls.
{
// MIN_LOD
tcu::TestCaseGroup* minLodGroup = new tcu::TestCaseGroup(m_testCtx, "min_lod", "Lod control: min lod");
group2D->addChild(minLodGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
minLodGroup->addChild(new Texture2DMinLodCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// MAX_LOD
tcu::TestCaseGroup* maxLodGroup = new tcu::TestCaseGroup(m_testCtx, "max_lod", "Lod control: max lod");
group2D->addChild(maxLodGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
maxLodGroup->addChild(new Texture2DMaxLodCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// BASE_LEVEL
tcu::TestCaseGroup* baseLevelGroup = new tcu::TestCaseGroup(m_testCtx, "base_level", "Base level");
group2D->addChild(baseLevelGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
baseLevelGroup->addChild(new Texture2DBaseLevelCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// MAX_LEVEL
tcu::TestCaseGroup* maxLevelGroup = new tcu::TestCaseGroup(m_testCtx, "max_level", "Max level");
group2D->addChild(maxLevelGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
maxLevelGroup->addChild(new Texture2DMaxLevelCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
}
// Cubemap cases.
for (int coordType = 0; coordType < DE_LENGTH_OF_ARRAY(cubeCoordTypes); coordType++)
{
tcu::TestCaseGroup* coordTypeGroup = new tcu::TestCaseGroup(m_testCtx, cubeCoordTypes[coordType].name, cubeCoordTypes[coordType].desc);
groupCube->addChild(coordTypeGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
{
coordTypeGroup->addChild(new TextureCubeMipmapCase(m_testCtx, m_context.getRenderContext(), m_context.getContextInfo(),
minFilterModes[minFilter].name, "",
cubeCoordTypes[coordType].type,
minFilterModes[minFilter].mode,
GL_CLAMP_TO_EDGE,
GL_CLAMP_TO_EDGE,
GL_RGBA, GL_UNSIGNED_BYTE, cubeMapSize));
}
}
// Cubemap mipmap generation variants.
{
tcu::TestCaseGroup* genMipmapGroup = new tcu::TestCaseGroup(m_testCtx, "generate", "Mipmap generation tests");
groupCube->addChild(genMipmapGroup);
for (int format = 0; format < DE_LENGTH_OF_ARRAY(formats); format++)
{
for (int hint = 0; hint < DE_LENGTH_OF_ARRAY(genHints); hint++)
{
std::ostringstream name;
name << formats[format].name
<< "_" << genHints[hint].name;
genMipmapGroup->addChild(new TextureCubeGenMipmapCase(m_testCtx, m_context.getRenderContext(), name.str().c_str(), "", formats[format].format, formats[format].dataType, genHints[hint].hint, cubeMapSize));
}
}
}
// Cubemap LOD controls.
{
// MIN_LOD
tcu::TestCaseGroup* minLodGroup = new tcu::TestCaseGroup(m_testCtx, "min_lod", "Lod control: min lod");
groupCube->addChild(minLodGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
minLodGroup->addChild(new TextureCubeMinLodCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// MAX_LOD
tcu::TestCaseGroup* maxLodGroup = new tcu::TestCaseGroup(m_testCtx, "max_lod", "Lod control: max lod");
groupCube->addChild(maxLodGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
maxLodGroup->addChild(new TextureCubeMaxLodCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// BASE_LEVEL
tcu::TestCaseGroup* baseLevelGroup = new tcu::TestCaseGroup(m_testCtx, "base_level", "Base level");
groupCube->addChild(baseLevelGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
baseLevelGroup->addChild(new TextureCubeBaseLevelCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// MAX_LEVEL
tcu::TestCaseGroup* maxLevelGroup = new tcu::TestCaseGroup(m_testCtx, "max_level", "Max level");
groupCube->addChild(maxLevelGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
maxLevelGroup->addChild(new TextureCubeMaxLevelCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
}
// 3D cases.
for (int coordType = 0; coordType < DE_LENGTH_OF_ARRAY(coordTypes); coordType++)
{
tcu::TestCaseGroup* coordTypeGroup = new tcu::TestCaseGroup(m_testCtx, coordTypes[coordType].name, coordTypes[coordType].desc);
group3D->addChild(coordTypeGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
{
for (int wrapMode = 0; wrapMode < DE_LENGTH_OF_ARRAY(wrapModes); wrapMode++)
{
// Add other size variants to basic cases only.
int sizeEnd = coordTypes[coordType].type == COORDTYPE_BASIC ? DE_LENGTH_OF_ARRAY(tex3DSizes) : 1;
for (int size = 0; size < sizeEnd; size++)
{
std::ostringstream name;
name << minFilterModes[minFilter].name
<< "_" << wrapModes[wrapMode].name;
if (tex3DSizes[size].name)
name << "_" << tex3DSizes[size].name;
coordTypeGroup->addChild(new Texture3DMipmapCase(m_context,
name.str().c_str(), "",
coordTypes[coordType].type,
minFilterModes[minFilter].mode,
wrapModes[wrapMode].mode,
wrapModes[wrapMode].mode,
wrapModes[wrapMode].mode,
GL_RGBA8,
tex3DSizes[size].width, tex3DSizes[size].height, tex3DSizes[size].depth));
}
}
}
}
// 3D bias variants.
{
tcu::TestCaseGroup* biasGroup = new tcu::TestCaseGroup(m_testCtx, "bias", "User-supplied bias value");
group3D->addChild(biasGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
biasGroup->addChild(new Texture3DMipmapCase(m_context,
minFilterModes[minFilter].name, "",
COORDTYPE_BASIC_BIAS,
minFilterModes[minFilter].mode,
GL_REPEAT, GL_REPEAT, GL_REPEAT,
GL_RGBA8,
tex3DSizes[0].width, tex3DSizes[0].height, tex3DSizes[0].depth));
}
// 3D LOD controls.
{
// MIN_LOD
tcu::TestCaseGroup* minLodGroup = new tcu::TestCaseGroup(m_testCtx, "min_lod", "Lod control: min lod");
group3D->addChild(minLodGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
minLodGroup->addChild(new Texture3DMinLodCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// MAX_LOD
tcu::TestCaseGroup* maxLodGroup = new tcu::TestCaseGroup(m_testCtx, "max_lod", "Lod control: max lod");
group3D->addChild(maxLodGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
maxLodGroup->addChild(new Texture3DMaxLodCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// BASE_LEVEL
tcu::TestCaseGroup* baseLevelGroup = new tcu::TestCaseGroup(m_testCtx, "base_level", "Base level");
group3D->addChild(baseLevelGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
baseLevelGroup->addChild(new Texture3DBaseLevelCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
// MAX_LEVEL
tcu::TestCaseGroup* maxLevelGroup = new tcu::TestCaseGroup(m_testCtx, "max_level", "Max level");
group3D->addChild(maxLevelGroup);
for (int minFilter = 0; minFilter < DE_LENGTH_OF_ARRAY(minFilterModes); minFilter++)
maxLevelGroup->addChild(new Texture3DMaxLevelCase(m_context, minFilterModes[minFilter].name, "", minFilterModes[minFilter].mode));
}
}
} // Functional
} // gles3
} // deqp