[PATCH] D21562: [AMDGPU] Wave and register controls

[Matt Arsenault via llvm-commits]

arsenm added inline comments.

================
Comment at: lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp:525
@@ +524,3 @@
+  ProgInfo.SGPRBlocks = ProgInfo.SGPRBlocks / RI->getSGPRAllocGranule() - 1;
+  // VGPRBlocks is actual number of VGPR blocks minus 1.
+  ProgInfo.VGPRBlocks = alignTo(ProgInfo.NumVGPRsForNumActiveWavesPerEU,
----------------
Line before comment

================
Comment at: lib/Target/AMDGPU/AMDGPUAsmPrinter.h:79
@@ +78,3 @@
+    uint32_t NumSGPRsForNumActiveWavesPerEU;
+    // Number of VGPRs that meets number of active waves per execution unit
+    // request.
----------------
Line before comment

================
Comment at: lib/Target/AMDGPU/AMDGPUSubtarget.cpp:211
@@ +210,3 @@
+  // Default minimum/maximum number of active waves per execution unit.
+  std::pair<unsigned, unsigned> Default = std::pair<unsigned, unsigned>(1, 0);
+
----------------
You can use those as the initializer value instead of the =

================
Comment at: lib/Target/AMDGPU/SIMachineFunctionInfo.h:359
@@ -354,4 +358,3 @@
 
-  /// \returns Number of reserved VGPRs for debugger usage.
-  unsigned getDebuggerReservedVGPRCount() const {
-    return DebuggerReservedVGPRCount;
+  /// Returns a pair of default/requested minimum/maximum flat work group sizes
+  /// for this function.
----------------
The Returns should be \returns to doxygenify, same for the rest of these

================
Comment at: lib/Target/AMDGPU/SIRegisterInfo.h:192
@@ +191,3 @@
+
+  /// Returns total number of SGPRs supported by the subtarget.
+  unsigned getTotalNumSGPRs(const SISubtarget &ST) const;
----------------
More \returns


https://reviews.llvm.org/D21562