Query API: Timestamp Query on Metal

- Add WriteTimestamp and ResolveQuerySet on Metal
- Enable end2end tests of timestamp query on Metal
- Lazy initialize the distination buffer in ResolveQuerySet
- Update part of end2end tests to test from render pass instead of
  command encoder due to we cannot write timestamp on CommandEncoder
  without any copy commands on Metal.

Bug: dawn:434
Change-Id: Ie9217e1f5a00bc252d6293ef7521c2e343ba9259
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/29020
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Jiawei Shao <jiawei.shao@intel.com>
Commit-Queue: Hao Li <hao.x.li@intel.com>

src/dawn_native/metal/CommandBufferMTL.mm

@@ -23,6 +23,7 @@
 #include "dawn_native/metal/ComputePipelineMTL.h"
 #include "dawn_native/metal/DeviceMTL.h"
 #include "dawn_native/metal/PipelineLayoutMTL.h"
+#include "dawn_native/metal/QuerySetMTL.h"
 #include "dawn_native/metal/RenderPipelineMTL.h"
 #include "dawn_native/metal/SamplerMTL.h"
 #include "dawn_native/metal/TextureMTL.h"
@@ -736,11 +737,39 @@ namespace dawn_native { namespace metal {
                 }
 
                 case Command::ResolveQuerySet: {
-                    return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation.");
+                    ResolveQuerySetCmd* cmd = mCommands.NextCommand<ResolveQuerySetCmd>();
+                    QuerySet* querySet = ToBackend(cmd->querySet.Get());
+                    Buffer* destination = ToBackend(cmd->destination.Get());
+
+                    destination->EnsureDataInitializedAsDestination(
+                        commandContext, cmd->destinationOffset, cmd->queryCount * sizeof(uint64_t));
+
+                    if (@available(macos 10.15, iOS 14.0, *)) {
+                        [commandContext->EnsureBlit()
+                              resolveCounters:querySet->GetCounterSampleBuffer()
+                                      inRange:NSMakeRange(cmd->firstQuery,
+                                                          cmd->firstQuery + cmd->queryCount)
+                            destinationBuffer:destination->GetMTLBuffer()
+                            destinationOffset:NSUInteger(cmd->destinationOffset)];
+                    } else {
+                        UNREACHABLE();
+                    }
+                    break;
                 }
 
                 case Command::WriteTimestamp: {
-                    return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation.");
+                    WriteTimestampCmd* cmd = mCommands.NextCommand<WriteTimestampCmd>();
+                    QuerySet* querySet = ToBackend(cmd->querySet.Get());
+
+                    if (@available(macos 10.15, iOS 14.0, *)) {
+                        [commandContext->EnsureBlit()
+                            sampleCountersInBuffer:querySet->GetCounterSampleBuffer()
+                                     atSampleIndex:NSUInteger(cmd->queryIndex)
+                                       withBarrier:YES];
+                    } else {
+                        UNREACHABLE();
+                    }
+                    break;
                 }
 
                 case Command::InsertDebugMarker: {
@@ -872,7 +901,17 @@ namespace dawn_native { namespace metal {
                 }
 
                 case Command::WriteTimestamp: {
-                    return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation.");
+                    WriteTimestampCmd* cmd = mCommands.NextCommand<WriteTimestampCmd>();
+                    QuerySet* querySet = ToBackend(cmd->querySet.Get());
+
+                    if (@available(macos 10.15, iOS 14.0, *)) {
+                        [encoder sampleCountersInBuffer:querySet->GetCounterSampleBuffer()
+                                          atSampleIndex:NSUInteger(cmd->queryIndex)
+                                            withBarrier:YES];
+                    } else {
+                        UNREACHABLE();
+                    }
+                    break;
                 }
 
                 default: {
@@ -1263,7 +1302,17 @@ namespace dawn_native { namespace metal {
                 }
 
                 case Command::WriteTimestamp: {
-                    return DAWN_UNIMPLEMENTED_ERROR("Waiting for implementation.");
+                    WriteTimestampCmd* cmd = mCommands.NextCommand<WriteTimestampCmd>();
+                    QuerySet* querySet = ToBackend(cmd->querySet.Get());
+
+                    if (@available(macos 10.15, iOS 14.0, *)) {
+                        [encoder sampleCountersInBuffer:querySet->GetCounterSampleBuffer()
+                                          atSampleIndex:NSUInteger(cmd->queryIndex)
+                                            withBarrier:YES];
+                    } else {
+                        UNREACHABLE();
+                    }
+                    break;
                 }
 
                 default: {

src/tests/end2end/BufferZeroInitTests.cpp

@@ -1120,7 +1120,7 @@ TEST_P(BufferZeroInitTest, SetIndexBuffer) {
     }
 }
 
-// Test the buffer will be lazily intialized correctly when its first use is an indirect buffer for
+// Test the buffer will be lazily initialized correctly when its first use is an indirect buffer for
 // DrawIndirect.
 TEST_P(BufferZeroInitTest, IndirectBufferForDrawIndirect) {
     // Bind the whole buffer as an indirect buffer.
@@ -1136,7 +1136,7 @@ TEST_P(BufferZeroInitTest, IndirectBufferForDrawIndirect) {
     }
 }
 
-// Test the buffer will be lazily intialized correctly when its first use is an indirect buffer for
+// Test the buffer will be lazily initialized correctly when its first use is an indirect buffer for
 // DrawIndexedIndirect.
 TEST_P(BufferZeroInitTest, IndirectBufferForDrawIndexedIndirect) {
     // Bind the whole buffer as an indirect buffer.
@@ -1152,7 +1152,7 @@ TEST_P(BufferZeroInitTest, IndirectBufferForDrawIndexedIndirect) {
     }
 }
 
-// Test the buffer will be lazily intialized correctly when its first use is an indirect buffer for
+// Test the buffer will be lazily initialized correctly when its first use is an indirect buffer for
 // DispatchIndirect.
 TEST_P(BufferZeroInitTest, IndirectBufferForDispatchIndirect) {
     // Bind the whole buffer as an indirect buffer.
@@ -1168,13 +1168,14 @@ TEST_P(BufferZeroInitTest, IndirectBufferForDispatchIndirect) {
     }
 }
 
-// Test the buffer will be lazily intialized correctly when its first use is in resolveQuerySet
+// Test the buffer will be lazily initialized correctly when its first use is in resolveQuerySet
 TEST_P(BufferZeroInitTest, ResolveQuerySet) {
     // Timestamp query is not supported on OpenGL
     DAWN_SKIP_TEST_IF(IsOpenGL());
 
-    // TODO(hao.x.li@intel.com): Remove it after timestamp query is implementated on Metal
+    // TODO(hao.x.li@intel.com): Crash occurs if we only call WriteTimestamp in a command encoder
-    DAWN_SKIP_TEST_IF(IsMetal());
+    // without any copy commands on Metal on AMD GPU. See https://crbug.com/dawn/545.
+    DAWN_SKIP_TEST_IF(IsMetal() && IsAMD());
 
     // Skip if timestamp extension is not supported on device
     DAWN_SKIP_TEST_IF(!SupportsExtensions({"timestamp_query"}));

src/tests/end2end/QueryTests.cpp

@@ -141,8 +141,9 @@ TEST_P(TimestampQueryTests, QuerySetCreation) {
 
 // Test calling timestamp query from command encoder
 TEST_P(TimestampQueryTests, TimestampOnCommandEncoder) {
-    // TODO(hao.x.li@intel.com): Waiting for timestamp query implementation on Metal
+    // TODO(hao.x.li@intel.com): Crash occurs if we only call WriteTimestamp in a command encoder
-    DAWN_SKIP_TEST_IF(IsMetal());
+    // without any copy commands on Metal on AMD GPU. See https://crbug.com/dawn/545.
+    DAWN_SKIP_TEST_IF(IsMetal() && IsAMD());
 
     constexpr uint32_t kQueryCount = 2;
 
@@ -161,9 +162,6 @@ TEST_P(TimestampQueryTests, TimestampOnCommandEncoder) {
 
 // Test calling timestamp query from render pass encoder
 TEST_P(TimestampQueryTests, TimestampOnRenderPass) {
-    // TODO(hao.x.li@intel.com): Waiting for timestamp query implementation on Metal
-    DAWN_SKIP_TEST_IF(IsMetal());
-
     constexpr uint32_t kQueryCount = 2;
 
     wgpu::QuerySet querySet = CreateQuerySetForTimestamp(kQueryCount);
@@ -184,9 +182,6 @@ TEST_P(TimestampQueryTests, TimestampOnRenderPass) {
 
 // Test calling timestamp query from compute pass encoder
 TEST_P(TimestampQueryTests, TimestampOnComputePass) {
-    // TODO(hao.x.li@intel.com): Waiting for timestamp query implementation on Metal
-    DAWN_SKIP_TEST_IF(IsMetal());
-
     constexpr uint32_t kQueryCount = 2;
 
     wgpu::QuerySet querySet = CreateQuerySetForTimestamp(kQueryCount);
@@ -206,11 +201,14 @@ TEST_P(TimestampQueryTests, TimestampOnComputePass) {
 
 // Test resolving timestamp query to one slot in the buffer
 TEST_P(TimestampQueryTests, ResolveToBufferWithOffset) {
-    // TODO(hao.x.li@intel.com): Failed on old Intel Vulkan driver on Windows, need investigation.
+    // TODO(hao.x.li@intel.com): Fail to resolve query to buffer with offset on Windows Vulkan and
+    // Metal on Intel platforms, need investigation.
     DAWN_SKIP_TEST_IF(IsWindows() && IsIntel() && IsVulkan());
+    DAWN_SKIP_TEST_IF(IsIntel() && IsMetal());
 
-    // TODO(hao.x.li@intel.com): Waiting for timestamp query implementation on Metal
+    // TODO(hao.x.li@intel.com): Crash occurs if we only call WriteTimestamp in a command encoder
-    DAWN_SKIP_TEST_IF(IsMetal());
+    // without any copy commands on Metal on AMD GPU. See https://crbug.com/dawn/545.
+    DAWN_SKIP_TEST_IF(IsMetal() && IsAMD());
 
     constexpr uint32_t kQueryCount = 2;
     constexpr uint64_t kZero = 0;