Handle denorm flush to negative zero in unpackhalf2x16 tests.

Whether input fp16 values are flushed to zero is implementation
dependent. However, negative denorm values can be flushed to negative
zero and the logic did not handle that case.

Affects: dEQP-VK.glsl.builtin.function.pack_unpack.unpackhalf2x16*
Component: Vulkan
VK-GL-CTS issue: 1341

Change-Id: I2cd4a68cb5e498cd714a7e55b27ae4bc05ccde2d

diff --git a/external/vulkancts/modules/vulkan/shaderexecutor/vktShaderPackingFunctionTests.cpp b/external/vulkancts/modules/vulkan/shaderexecutor/vktShaderPackingFunctionTests.cpp
index 1d4dd8a..469369d 100644 (file)
--- a/external/vulkancts/modules/vulkan/shaderexecutor/vktShaderPackingFunctionTests.cpp
+++ b/external/vulkancts/modules/vulkan/shaderexecutor/vktShaderPackingFunctionTests.cpp
@@ -856,7 +856,7 @@ private:
 
                if (conversion != CONVERTED && denorm)
                {
-                       if (resBits == 0)
+                       if (resBits == 0 || (ref < 0 && resBits == 0x80000000UL))
                        {
                                conversion = ZERO_FLUSHED;
                                return DE_TRUE;