PIX SM 6.6 resource access tracking (#3594)

This change updates the existing PIX resource-tracking code to handle dynamic resources (and SM6.6's resource binding apparatus in general).
The output UAV is now segmented into three parts: the original formatted buffer at the beginning, for old-style createHandle resources, followed by a block for resource (texture, buffer etc) access, followed by a block for sampler access.
The latter two are divided into 8-byte records. The first dword records writes to a resource/sampler, the second reads.
The writes are encoded bit fields denoting the access performed by the shader.

(cherry picked from commit 880c135)

Author: jeffnn

lib/DxilPIXPasses/DxilShaderAccessTracking.cpp

@@ -1,4 +1,4 @@
-// RUN: %dxc -ECSMain -Tcs_6_0 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=S0:1:1i1;U0:2:10i0;.. | %FileCheck %s
+// RUN: %dxc -ECSMain -Tcs_6_0 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=S0:1:1i1;U0:2:10i0;.0;0;0. | %FileCheck %s
 
 // Check we added the UAV:
 // CHECK:  %PIX_CountUAV_Handle = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 1, i32 0, i1 false)

tools/clang/test/HLSLFileCheck/pix/AccessTracking.hlsl

@@ -1,4 +1,4 @@
-// RUN: %dxc -Emain -Tcs_6_5 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=M0:0:1i0;S0:1:1i0;U0:2:10i0;.. | %FileCheck %s
+// RUN: %dxc -Emain -Tcs_6_5 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=M0:0:1i0;S0:1:1i0;U0:2:10i0;.0;0;0. | %FileCheck %s
 
 // Check we added the UAV:
 // CHECK:  %PIX_CountUAV_Handle = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 1, i32 0, i1 false)

tools/clang/test/HLSLFileCheck/pix/AccessTrackingForSamplerFeedback.hlsl

@@ -0,0 +1,26 @@
+// RUN: %dxc -EMain -Tcs_6_6 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=.256;512;1024. | %FileCheck %s
+
+static RWByteAddressBuffer DynamicBuffer = ResourceDescriptorHeap[1];
+[numthreads(1, 1, 1)]
+void Main()
+{
+    uint val = DynamicBuffer.Load(0u);
+    DynamicBuffer.Store(0u, val);
+}
+
+// check it's 6.6:
+// CHECK: call %dx.types.Handle @dx.op.createHandleFromBinding
+
+// Check we wrote to the PIX UAV:
+// CHECK: call void @dx.op.bufferStore.i32
+
+// Offset for buffer Load should be 256 + 8 (skip out-of-bounds record) + 8 (it's the 1th resource) + 4 (offset to the "read" field) = 276
+// The large integer is encoded flags for the ResourceAccessStyle (an enumerated type in lib\DxilPIXPasses\DxilShaderAccessTracking.cpp) for this access
+// CHECK:i32 276, i32 undef, i32 1375731712
+// CHECK:rawBufferLoad
+
+// Offset for buffer Store should be 256 + 8 (skip out-of-bounds record) + 8 (it's the 1th resource) + 0 (offset to the "write" field) = 272
+// The large integer is encoded flags for the ResourceAccessStyle (an enumerated type in lib\DxilPIXPasses\DxilShaderAccessTracking.cpp) for this access
+// CHECK:i32 272, i32 undef, i32 1392508928
+// CHECK:rawBufferStore
+

tools/clang/test/HLSLFileCheck/pix/DynamicResourceAccessTracking.hlsl

@@ -0,0 +1,19 @@
+// RUN: %dxc -EMain -Tcs_6_6 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=.256;272;1024. | %FileCheck %s
+
+static RWByteAddressBuffer DynamicBuffer = ResourceDescriptorHeap[1];
+[numthreads(1, 1, 1)]
+void Main()
+{
+    uint val = DynamicBuffer.Load(0u);
+    DynamicBuffer.Store(0u, val);
+}
+
+// check it's 6.6:
+// CHECK: call %dx.types.Handle @dx.op.createHandleFromBinding
+
+// The start of resource records has been passed in as 256. The limit of resource records is 272. 272-256 = 16.
+// 8 bytes per record means we have one record for out-of-bounds (that comes first), and one record for resource index 0.
+// The above HLSL references resource descriptor 1, so is out-of-bounds. Offset for out-of-bounds should thus be 256:
+// The large integer is encoded flags for the ResourceAccessStyle (an enumerated type in lib\DxilPIXPasses\DxilShaderAccessTracking.cpp) for this access
+// CHECK:i32 256, i32 undef, i32 1375731712
+// CHECK:rawBufferLoad

tools/clang/test/HLSLFileCheck/pix/DynamicResourceOutOfBounds.hlsl

@@ -0,0 +1,26 @@
+// RUN: %dxc -EMain -Tps_6_6 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=S0:1:1i1;.256;512;1024. | %FileCheck %s
+
+static sampler sampler0 = SamplerDescriptorHeap[0];
+static sampler sampler3 = SamplerDescriptorHeap[3];
+Texture2D tx : register(t2);
+
+float4 Main() : SV_Target
+{
+    float4 a = tx.Sample(sampler0, float2(0,0));
+    float4 b = tx.Sample(sampler3, float2(0,0));
+    return a + b;
+}
+
+// check it's 6.6:
+// CHECK: call %dx.types.Handle @dx.op.createHandleFromBinding
+
+// The large integers are encoded flags for the ResourceAccessStyle (an enumerated type in lib\DxilPIXPasses\DxilShaderAccessTracking.cpp) for this access
+
+// Check we wrote sampler data to the PIX UAV. We told the pass to output starting at offset 512.
+// Add 8 to skip the "out of bounds" record. Add 4 to point to the "read" field within the next entry = 524
+// CHECK: call void @dx.op.bufferStore.i32(
+// CHECK:i32 524, i32 undef, i32 16777216
+
+// twice: 512 + 8 + 8*3+4 = 548
+// CHECK: call void @dx.op.bufferStore.i32(
+// CHECK:i32 548, i32 undef, i32 16777216

tools/clang/test/HLSLFileCheck/pix/DynamicSamplerAccessTracking.hlsl

@@ -0,0 +1,24 @@
+// RUN: %dxc -EMain -Tps_6_6 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=S0:1:1i1;.256;512;520. | %FileCheck %s
+
+static sampler sampler0 = SamplerDescriptorHeap[0];
+static sampler sampler3 = SamplerDescriptorHeap[3];
+Texture2D tx : register(t2);
+
+float4 Main() : SV_Target
+{
+    float4 a = tx.Sample(sampler0, float2(0,0));
+    float4 b = tx.Sample(sampler3, float2(0,0));
+    return a + b;
+}
+
+// check it's 6.6:
+// CHECK: call %dx.types.Handle @dx.op.createHandleFromBinding
+
+// The large integers are encoded flags for the ResourceAccessStyle (an enumerated type in lib\DxilPIXPasses\DxilShaderAccessTracking.cpp) for this access
+
+// The start of sampler records has been passed in as 512. The limit of the whole buffer is 520, leaving just one eight-byte record for the out-of-bounds record.
+// There are therefore no expected in-bounds references to samplers, so any such reference should go to the out-of-bounds offset at 512:
+
+// Out of bounds sampler access should be at offset 512
+// CHECK: call void @dx.op.bufferStore.i32(
+// CHECK:i32 512, i32 undef, i32 16777216

tools/clang/test/HLSLFileCheck/pix/DynamicSamplerOutOfBounds.hlsl

@@ -1,4 +1,4 @@
-// RUN: %dxc -T vs_6_5 -E main %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=S0:1:1i1;U0:2:10i0;.. | FileCheck %s
+// RUN: %dxc -T vs_6_5 -E main %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=S0:1:1i1;U0:2:10i0;.0;0;0. | FileCheck %s
 
 // CHECK: call void @dx.op.rayQuery_TraceRayInline
 // CHECK: call void @dx.op.bufferStore.i32(i32 69, %dx.types.Handle

tools/clang/test/HLSLFileCheck/pix/TraceRayInline.hlsl

@@ -1,4 +1,4 @@
-// RUN: %dxc -enable-16bit-types -Emain -Tcs_6_3 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=U0:2:10i0;.. | %FileCheck %s
+// RUN: %dxc -enable-16bit-types -Emain -Tcs_6_3 %s | %opt -S -hlsl-dxil-pix-shader-access-instrumentation,config=U0:2:10i0;.0;0;0. | %FileCheck %s
 
 // Check that the expected PIX UAV read-tracking is emitted (the atomicBinOp "|= 1") followed by the expected raw read:
 

tools/clang/test/HLSLFileCheck/pix/rawBufferStore.hlsl

@@ -202,6 +202,7 @@ class CompilerTest : public ::testing::Test {
   TEST_METHOD(BatchPasses)
   TEST_METHOD(BatchShaderTargets)
   TEST_METHOD(BatchValidation)
+  TEST_METHOD(BatchPIX)
 
   TEST_METHOD(SubobjectCodeGenErrors)
   BEGIN_TEST_METHOD(ManualFileCheckTest)
@@ -468,10 +469,20 @@ class CompilerTest : public ::testing::Test {
         dumpStr = dumpPath + (wRelPath.m_psz + suitePath.size());
       }
 
-      WEX::Logging::Log::StartGroup(wRelPath);
+      class ScopedLogGroup
+      {
+        LPWSTR m_path;
+
+      public:
+          ScopedLogGroup(LPWSTR path)
+          : m_path(path)
+          { WEX::Logging::Log::StartGroup(m_path); }
+          ~ScopedLogGroup() { WEX::Logging::Log::EndGroup(m_path); }
+      };
+
+      ScopedLogGroup cleanup(wRelPath);
       CodeGenTestCheck(wRelPath, /*implicitDir*/ false,
         dumpStr.empty() ? nullptr : dumpStr.c_str());
-      WEX::Logging::Log::EndGroup(wRelPath);
 
       numTestsRun++;
     }
@@ -3138,6 +3149,10 @@ TEST_F(CompilerTest, BatchValidation) {
   CodeGenTestCheckBatchDir(L"validation");
 }
 
+TEST_F(CompilerTest, BatchPIX) {
+  CodeGenTestCheckBatchDir(L"PIX");
+}
+
 TEST_F(CompilerTest, BatchSamples) {
   CodeGenTestCheckBatchDir(L"samples");
 }