Reinstate memory barriers on non-Apple GPUs.

Ensure non-Apple GPU's enable memory barriers.
A previous commit inadvertently disabled GPU memory barriers.

Change tests for memory barriers to runtime test for Apple GPU, instead of
build-time test for Apple Silicon, to accommodate running on Rosetta2, and
refactor tests for Apple Silicon and OS version on some macOS GPU feature settings.

Docs/Whats_New.md

@@ -18,6 +18,7 @@ MoltenVK 1.1.9
 
 Released TBD
 
+- Reinstate memory barriers on non-Apple GPUs, which were inadvertently disabled in an earlier update.
 - Update *glslang* version, to use `python3` in *glslang* scripts, to replace missing `python` on *macOS 12.3*.
 - Fix alignment between outputs and inputs between shader stages when using nested structures.
 - Fix issue where the depth component of a stencil-only renderpass attachment was incorrectly attempting to be stored.

MoltenVK/MoltenVK/Commands/MVKCmdPipeline.mm

@@ -113,14 +113,13 @@ void MVKCmdPipelineBarrier<N>::encode(MVKCommandEncoder* cmdEncoder) {
 	}
 #endif
 
-#if MVK_APPLE_SILICON
 	// Apple GPUs do not support renderpass barriers, and do not support rendering/writing
 	// to an attachment and then reading from that attachment within a single renderpass.
 	// So, in the case where we are inside a Metal renderpass, we need to split those activities
 	// into separate Metal renderpasses. Since this is a potentially expensive operation,
 	// verify that at least one attachment is being used both as an input and render attachment
 	// by checking for a VK_IMAGE_LAYOUT_GENERAL layout.
-	if (cmdEncoder->_mtlRenderEncoder) {
+	if (cmdEncoder->_mtlRenderEncoder && cmdEncoder->getDevice()->_pMetalFeatures->tileBasedDeferredRendering) {
 		bool needsRenderpassRestart = false;
 		for (auto& b : _barriers) {
 			if (b.type == MVKPipelineBarrier::Image && b.newLayout == VK_IMAGE_LAYOUT_GENERAL) {
@@ -133,7 +132,6 @@ void MVKCmdPipelineBarrier<N>::encode(MVKCommandEncoder* cmdEncoder) {
 			cmdEncoder->beginMetalRenderPass(kMVKCommandUseRestartSubpass);
 		}
 	}
-#endif
 
 	MVKDevice* mvkDvc = cmdEncoder->getDevice();
 	MVKCommandUse cmdUse = kMVKCommandUsePipelineBarrier;

MoltenVK/MoltenVK/Commands/MVKCmdTransfer.mm

@@ -1310,22 +1310,22 @@ void MVKCmdClearAttachments<N>::encode(MVKCommandEncoder* cmdEncoder) {
     [mtlRendEnc drawPrimitives: MTLPrimitiveTypeTriangle vertexStart: 0 vertexCount: vtxCnt];
     [mtlRendEnc popDebugGroup];
 
-#if MVK_APPLE_SILICON
 	// Apple GPUs do not support rendering/writing to an attachment and then reading from
 	// that attachment within a single Metal renderpass. So, if any of the attachments just
 	// cleared is an input attachment, we need to restart into separate Metal renderpasses.
-	bool needsRenderpassRestart = false;
-	for (uint32_t caIdx = 0; caIdx < caCnt; caIdx++) {
-		if (_rpsKey.isAttachmentEnabled(caIdx) && subpass->isColorAttachmentAlsoInputAttachment(caIdx)) {
-			needsRenderpassRestart = true;
-			break;
+	if (cmdEncoder->getDevice()->_pMetalFeatures->tileBasedDeferredRendering) {
+		bool needsRenderpassRestart = false;
+		for (uint32_t caIdx = 0; caIdx < caCnt; caIdx++) {
+			if (_rpsKey.isAttachmentEnabled(caIdx) && subpass->isColorAttachmentAlsoInputAttachment(caIdx)) {
+				needsRenderpassRestart = true;
+				break;
+			}
+		}
+		if (needsRenderpassRestart) {
+			cmdEncoder->encodeStoreActions(true);
+			cmdEncoder->beginMetalRenderPass(kMVKCommandUseRestartSubpass);
 		}
 	}
-	if (needsRenderpassRestart) {
-		cmdEncoder->encodeStoreActions(true);
-		cmdEncoder->beginMetalRenderPass(kMVKCommandUseRestartSubpass);
-	}
-#endif
 
 	// Return to the previous rendering state on the next render activity
 	cmdEncoder->_graphicsPipelineState.markDirty();

MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm

@@ -1478,38 +1478,42 @@ void MVKPhysicalDevice::initMetalFeatures() {
 	}
 
 #if MVK_XCODE_12
-	if ( mvkOSVersionIsAtLeast(10.16) ) {
+	if ( mvkOSVersionIsAtLeast(11.0) ) {
 		_metalFeatures.mslVersionEnum = MTLLanguageVersion2_3;
-		if (supportsMTLGPUFamily(Apple5)) {
-			// This is an Apple GPU--treat it accordingly.
-			_metalFeatures.mtlCopyBufferAlignment = 1;
-			_metalFeatures.mtlBufferAlignment = 16;     // Min float4 alignment for typical vertex buffers. MTLBuffer may go down to 4 bytes for other data.
-			_metalFeatures.maxQueryBufferSize = (64 * KIBI);
-			_metalFeatures.maxPerStageDynamicMTLBufferCount = _metalFeatures.maxPerStageBufferCount;
-			_metalFeatures.postDepthCoverage = true;
-			_metalFeatures.renderLinearTextures = true;
-			_metalFeatures.tileBasedDeferredRendering = true;
-			if (supportsMTLGPUFamily(Apple6)) {
-				_metalFeatures.astcHDRTextures = true;
-			}
-			if (supportsMTLGPUFamily(Apple7)) {
-				_metalFeatures.maxQueryBufferSize = (256 * KIBI);
-			}
-		}
-	} else
+	}
 #endif
-	{
+#if MVK_XCODE_13
+	if ( mvkOSVersionIsAtLeast(12.0) ) {
+		_metalFeatures.mslVersionEnum = MTLLanguageVersion2_4;
+	}
+#endif
+
+	// This is an Apple GPU--treat it accordingly.
+	if (supportsMTLGPUFamily(Apple1)) {
+		_metalFeatures.mtlCopyBufferAlignment = 1;
+		_metalFeatures.mtlBufferAlignment = 16;     // Min float4 alignment for typical vertex buffers. MTLBuffer may go down to 4 bytes for other data.
+		_metalFeatures.maxQueryBufferSize = (64 * KIBI);
+		_metalFeatures.maxPerStageDynamicMTLBufferCount = _metalFeatures.maxPerStageBufferCount;
+		_metalFeatures.postDepthCoverage = true;
+		_metalFeatures.renderLinearTextures = true;
+		_metalFeatures.tileBasedDeferredRendering = true;
+		if (supportsMTLGPUFamily(Apple6)) {
+			_metalFeatures.astcHDRTextures = true;
+		}
+		if (supportsMTLGPUFamily(Apple7)) {
+			_metalFeatures.maxQueryBufferSize = (256 * KIBI);
+		}
+	}
+
+	// Don't use barriers in render passes on Apple GPUs. Apple GPUs don't support them,
+	// and in fact Metal's validation layer will complain if you try to use them.
+	if ( !supportsMTLGPUFamily(Apple1) ) {
 		if (supportsMTLFeatureSet(macOS_GPUFamily1_v4)) {
 			_metalFeatures.memoryBarriers = true;
 		}
 		_metalFeatures.textureBarriers = true;
 	}
 
-#if MVK_XCODE_13
-	if ( mvkOSVersionIsAtLeast(12.0) ) {
-		_metalFeatures.mslVersionEnum = MTLLanguageVersion2_4;
-	}
-#endif
 #endif
 
     // Note the selector name, which is different from the property name.

MoltenVK/MoltenVK/GPUObjects/MVKRenderPass.mm

@@ -809,14 +809,12 @@ MVKRenderPassAttachment::MVKRenderPassAttachment(MVKRenderPass* renderPass,
 #pragma mark MVKRenderPass
 
 VkExtent2D MVKRenderPass::getRenderAreaGranularity() {
-#if MVK_APPLE_SILICON
     if (_device->_pMetalFeatures->tileBasedDeferredRendering) {
         // This is the tile area.
         // FIXME: We really ought to use MTLRenderCommandEncoder.tile{Width,Height}, but that requires
         // creating a command buffer.
         return { 32, 32 };
     }
-#endif
     return { 1, 1 };
 }