P4 to Git Change 1397729 by kzhuravl@kzhuravl-fiji-ocllc on 2017/04/12 20:05:08

SWDEV-116535 - Report CL_FP_DENORM for single fp config for gfx9 for LC for rocm/pal and force denorms on based on AMD_GPU_FORCE_SINGLE_FP_DENORM for rocm.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#23 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#47 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#62 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#16 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.hpp#7 edit

rocclr/runtime/device/pal/palsettings.cpp

@@ -414,6 +414,14 @@ Settings::create(
     resourceCacheSize_ = std::min(resourceCacheSize_, 512 * Mi);
 #endif
 
+#if defined(WITH_LIGHTNING_COMPILER)
+    switch (palProp.gfxLevel) {
+    case Pal::GfxIpLevel::GfxIp9:
+        singleFpDenorm_ = true;
+        break;
+    }
+#endif // WITH_LIGHTNING_COMPILER
+
     // Override current device settings
     override();
 

rocclr/runtime/device/rocm/rocdevice.cpp

@@ -99,7 +99,7 @@ bool NullDevice::create(const AMDDeviceInfo& deviceInfo) {
 
     settings_ = new Settings();
     roc::Settings* hsaSettings = static_cast<roc::Settings*>(settings_);
-    if ((hsaSettings == nullptr) || !hsaSettings->create(false)) {
+    if ((hsaSettings == nullptr) || !hsaSettings->create(false, deviceInfo_.gfxipVersion_)) {
             LogError("Error creating settings for nullptr HSA device");
             return false;
     }
@@ -618,7 +618,7 @@ Device::mapHSADeviceToOpenCLDevice(hsa_agent_t dev)
     settings_ = new Settings();
     roc::Settings* hsaSettings = static_cast<roc::Settings*>(settings_);
     if ((hsaSettings == nullptr) ||
-        !hsaSettings->create((agent_profile_ == HSA_PROFILE_FULL))) {
+        !hsaSettings->create((agent_profile_ == HSA_PROFILE_FULL), deviceInfo_.gfxipVersion_)) {
         return false;
     }
 
@@ -978,6 +978,11 @@ Device::populateOCLDeviceConstants()
         info_.doubleFPConfig_ = info_.singleFPConfig_ | CL_FP_DENORM;
         info_.singleFPConfig_ |= CL_FP_CORRECTLY_ROUNDED_DIVIDE_SQRT;
     }
+
+    if (hsa_settings->singleFpDenorm_) {
+        info_.singleFPConfig_ |= CL_FP_DENORM;
+    }
+
     info_.preferredPlatformAtomicAlignment_ = 0;
     info_.preferredGlobalAtomicAlignment_ = 0;
     info_.preferredLocalAtomicAlignment_ = 0;

rocclr/runtime/device/rocm/rocprogram.cpp

@@ -1397,7 +1397,8 @@ HSAILProgram::codegenOptions(amd::option::Options* options)
     std::string optionsStr;
 
 #if !defined(WITH_LIGHTNING_COMPILER)
-    if (dev().deviceInfo().gfxipVersion_ < 900) {
+    if (dev().deviceInfo().gfxipVersion_ < 900 ||
+            !dev().settings().singleFpDenorm_) {
         optionsStr.append(" -cl-denorms-are-zero");
     }
 #endif // !defined(WITH_LIGHTNING_COMPILER)

rocclr/runtime/device/rocm/rocsettings.cpp

@@ -68,10 +68,13 @@ Settings::Settings()
     const static size_t MaxPinnedXferSize = 32;
     pinnedXferSize_     = std::min(GPU_PINNED_XFER_SIZE, MaxPinnedXferSize) * Mi;
     pinnedMinXferSize_  = std::min(GPU_PINNED_MIN_XFER_SIZE * Ki, pinnedXferSize_);
+
+    // Don't support Denormals for single precision by default
+    singleFpDenorm_ = false;
 }
 
 bool
-Settings::create(bool fullProfile)
+Settings::create(bool fullProfile, int gfxipVersion)
 {
     customHostAllocator_ = false;
 
@@ -111,6 +114,14 @@ Settings::create(bool fullProfile)
     enableExtension(ClKhrDepthImages);
     supportDepthsRGB_ = true;
 
+#if defined(WITH_LIGHTNING_COMPILER)
+    switch (gfxipVersion) {
+    case 900:
+        singleFpDenorm_ = true;
+        break;
+    }
+#endif // WITH_LIGHTNING_COMPILER
+
     // Override current device settings
     override();
 
@@ -140,6 +151,19 @@ Settings::override()
     if (!flagIsDefault(GPU_PINNED_MIN_XFER_SIZE)) {
         pinnedMinXferSize_  = std::min(GPU_PINNED_MIN_XFER_SIZE * Ki, pinnedXferSize_);
     }
+
+    if (!flagIsDefault(AMD_GPU_FORCE_SINGLE_FP_DENORM)) {
+        switch (AMD_GPU_FORCE_SINGLE_FP_DENORM) {
+        case 0:
+            singleFpDenorm_ = false;
+            break;
+        case 1:
+            singleFpDenorm_ = true;
+            break;
+        default:
+            break;
+        }
+    }
 }
 
 } // namespace roc

rocclr/runtime/device/rocm/rocsettings.hpp

@@ -29,7 +29,8 @@ public:
             uint    imageDMA_: 1;               //!< Enable direct image DMA transfers
             uint    stagedXferRead_: 1;         //!< Uses a staged buffer read
             uint    stagedXferWrite_: 1;        //!< Uses a staged buffer write
-            uint    reserved_: 22;
+            uint    singleFpDenorm_: 1;         //!< Support Single FP Denorm
+            uint    reserved_: 21;
         };
         uint    value_;
     };
@@ -58,7 +59,7 @@ public:
     Settings();
 
     //! Creates settings
-    bool create(bool fullProfile);
+    bool create(bool fullProfile, int gfxipVersion);
 
 private:
     //! Disable copy constructor