P4 to Git Change 1136253 by gandryey@gera-w8 on 2015/03/31 12:51:46

ECR #304775 - Mipmaps support in OCL
	- Add changes for the sampler usage with mipmaped images

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_sampler.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#504 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.hpp#141 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp#113 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLDevice.h#46 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/sampler.hpp#7 edit

rocclr/runtime/device/gpu/gpudevice.cpp

@@ -1904,7 +1904,7 @@ Device::createSampler(const amd::Sampler& owner, device::Sampler** sampler) cons
     *sampler = NULL;
     if (settings().hsail_ || (settings().oclVersion_ >= OpenCL20)) {
         Sampler* gpuSampler = new Sampler(*this);
-        if ((NULL == gpuSampler) || !gpuSampler->create(owner.state())) {
+        if ((NULL == gpuSampler) || !gpuSampler->create(owner)) {
             delete gpuSampler;
             return false;
         }
@@ -2449,7 +2449,9 @@ Device::destroyScratchBuffers()
 }
 
 void
-Device::fillHwSampler(uint32_t state, void* hwState, uint32_t hwStateSize) const
+Device::fillHwSampler(
+    uint32_t state, void* hwState, uint32_t hwStateSize,
+    uint32_t mipFilter, float minLod, float maxLod) const
 {
     // All GSL sampler's parameters are in floats
     uint32_t    gslAddress = GSL_CLAMP_TO_BORDER;
@@ -2483,8 +2485,25 @@ Device::fillHwSampler(uint32_t state, void* hwState, uint32_t hwStateSize) const
         gslMagFilter = GSL_MAG_LINEAR;
     }
 
+    if (mipFilter == CL_FILTER_NEAREST) {
+        if (gslMinFilter == GSL_MIN_NEAREST) {
+            gslMinFilter = GSL_MIN_NEAREST_MIPMAP_NEAREST;
+        }
+        else {
+            gslMinFilter = GSL_MIN_LINEAR_MIPMAP_NEAREST;
+        }
+    }
+    else if (mipFilter == CL_FILTER_LINEAR) {
+        if (gslMinFilter == GSL_MIN_NEAREST) {
+            gslMinFilter = GSL_MIN_NEAREST_MIPMAP_LINEAR;
+        }
+        else {
+            gslMinFilter = GSL_MIN_LINEAR_MIPMAP_LINEAR;
+        }
+    }
+
     fillSamplerHwState(unnorm, gslMinFilter, gslMagFilter,
-        gslAddress, hwState, hwStateSize);
+        gslAddress, minLod, maxLod, hwState, hwStateSize);
 }
 
 void*
@@ -2564,8 +2583,7 @@ Device::SrdManager::~SrdManager()
 }
 
 bool
-Sampler::create(
-    uint32_t oclSamplerState)
+Sampler::create(uint32_t oclSamplerState)
 {
     hwSrd_ = dev_.srds().allocSrdSlot(&hwState_);
     if (0 == hwSrd_) {
@@ -2575,6 +2593,18 @@ Sampler::create(
     return true;
 }
 
+bool
+Sampler::create(const amd::Sampler& owner)
+{
+    hwSrd_ = dev_.srds().allocSrdSlot(&hwState_);
+    if (0 == hwSrd_) {
+        return false;
+    }
+    dev_.fillHwSampler(owner.state(), hwState_, HsaSamplerObjectSize,
+        owner.mipFilter(), owner.minLod(), owner.maxLod());
+    return true;
+}
+
 Sampler::~Sampler()
 {
     dev_.srds().freeSrdSlot(hwSrd_);

rocclr/runtime/device/gpu/gpudevice.hpp

@@ -140,6 +140,10 @@ class VirtualDevice;
 class PrintfDbg;
 class ThreadTrace;
 
+#ifndef CL_FILTER_NONE
+#define CL_FILTER_NONE 0x1142
+#endif
+
 class Sampler : public device::Sampler
 {
 public:
@@ -154,6 +158,11 @@ public:
         uint32_t oclSamplerState    //!< OCL sampler state
         );
 
+    //! Creates a device sampler from the OCL sampler state
+    bool create(
+        const amd::Sampler& owner   //!< AMD sampler object
+        );
+
     const void* hwState() const { return hwState_; }
 
 private:
@@ -518,9 +527,12 @@ public:
 
     //! Set GSL sampler to the specified state
     void    fillHwSampler(
-        uint32_t    state,              //!< Sampler's OpenCL state
-        void*       hwState,            //!< Sampler's HW state
-        uint32_t    hwStateSize         //!< Size of sampler's HW state
+        uint32_t    state,          //!< Sampler's OpenCL state
+        void*       hwState,        //!< Sampler's HW state
+        uint32_t    hwStateSize,    //!< Size of sampler's HW state
+        uint32_t    mipFilter = CL_FILTER_NONE, //!< Mip filter
+        float       minLod = 0.f,           //!< Min level of detail
+        float       maxLod = CL_MAXFLOAT    //!< Max level of detail
         ) const;
 
     //! host memory alloc

rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp

@@ -1726,7 +1726,8 @@ CALGSLDevice::fillImageHwState(gslMemObject mem, void* hwState, uint32 hwStateSi
 }
 
 void
-CALGSLDevice::fillSamplerHwState(bool unnorm, uint32 min, uint32 mag, uint32 addr, void* hwState, uint32 hwStateSize) const
+CALGSLDevice::fillSamplerHwState(bool unnorm, uint32 min, uint32 mag, uint32 addr,
+    float minLod, float maxLod, void* hwState, uint32 hwStateSize) const
 {
     amd::ScopedLock k(gslDeviceOps());
     m_textureSampler->setUnnormalizedMode(m_cs, unnorm);
@@ -1735,6 +1736,8 @@ CALGSLDevice::fillSamplerHwState(bool unnorm, uint32 min, uint32 mag, uint32 add
     m_textureSampler->setWrap(m_cs, GSL_TEXTURE_WRAP_S, static_cast<gslTexParameterParamWrap>(addr));
     m_textureSampler->setWrap(m_cs, GSL_TEXTURE_WRAP_T, static_cast<gslTexParameterParamWrap>(addr));
     m_textureSampler->setWrap(m_cs, GSL_TEXTURE_WRAP_R, static_cast<gslTexParameterParamWrap>(addr));
+    m_textureSampler->setMinLOD(m_cs, static_cast<float32>(minLod));
+    m_textureSampler->setMaxLOD(m_cs, static_cast<float32>(maxLod));
 
     m_textureSampler->getSamplerSrd(m_cs, hwState, hwStateSize);
 }

rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.h

@@ -126,7 +126,8 @@ public:
 
     void fillImageHwState(gslMemObject mem, void* hwState, uint32 hwStateSize) const;
 
-    void fillSamplerHwState(bool unnorm, uint32 min, uint32 mag, uint32 addr, void* hwState, uint32 hwStateSize) const;
+    void fillSamplerHwState(bool unnorm, uint32 min, uint32 mag, uint32 addr,
+        float minLod, float maxLod, void* hwState, uint32 hwStateSize) const;
 
     gslSamplerObject txSampler() const { return m_textureSampler; }