[llvm] AMDGPU: Remove wavefrontsize64 feature from dummy target (PR #117410)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/117410

This is a refinement for the existing hack. With this,
the default target will have neither wavefrontsize feature
present, unless it was explicitly specified. That is,
getWavefrontSize() == 64 no longer implies +wavefrontsize64.
getWavefrontSize() == 32 does imply +wavefrontsize32.

Continue to assume the value is 64 with no wavesize feature.
This maintains the codegenable property without any code
that directly cares about the wavesize needs to worry about it.

Introduce an isWaveSizeKnown helper to check if we know the
wavesize is accurate based on having one of the features explicitly
set, or a known target-cpu.

>From dad089846eba71cc8d9c8cbe96ea6a0d67900593 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Fri, 22 Nov 2024 17:21:18 -0800
Subject: [PATCH] AMDGPU: Remove wavefrontsize64 feature from dummy target

This is a refinement for the existing hack. With this,
the default target will have neither wavefrontsize feature
present, unless it was explicitly specified. That is,
getWavefrontSize() == 64 no longer implies +wavefrontsize64.
getWavefrontSize() == 32 does imply +wavefrontsize32.

Continue to assume the value is 64 with no wavesize feature.
This maintains the codegenable property without any code
that directly cares about the wavesize needs to worry about it.

Introduce an isWaveSizeKnown helper to check if we know the
wavesize is accurate based on having one of the features explicitly
set, or a known target-cpu.
---
 llvm/lib/Target/AMDGPU/GCNProcessors.td          |  4 ++--
 llvm/lib/Target/AMDGPU/GCNSubtarget.cpp          | 16 +++++++---------
 llvm/lib/Target/AMDGPU/GCNSubtarget.h            |  8 ++++++++
 .../AMDGPU/MCTargetDesc/AMDGPUInstPrinter.cpp    |  6 +++---
 4 files changed, 20 insertions(+), 14 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/GCNProcessors.td b/llvm/lib/Target/AMDGPU/GCNProcessors.td
index 3403cbab526d46..6241fa6e22ab8b 100644
--- a/llvm/lib/Target/AMDGPU/GCNProcessors.td
+++ b/llvm/lib/Target/AMDGPU/GCNProcessors.td
@@ -9,11 +9,11 @@
 // The code produced for "generic" is only useful for tests and cannot
 // reasonably be expected to execute on any particular target.
 def : ProcessorModel<"generic", NoSchedModel,
-  [FeatureWavefrontSize64, FeatureGDS, FeatureGWS]
+  [FeatureGDS, FeatureGWS]
 >;
 
 def : ProcessorModel<"generic-hsa", NoSchedModel,
-  [FeatureWavefrontSize64, FeatureGDS, FeatureGWS, FeatureFlatAddressSpace]
+  [FeatureGDS, FeatureGWS, FeatureFlatAddressSpace]
 >;
 
 //===------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.cpp b/llvm/lib/Target/AMDGPU/GCNSubtarget.cpp
index 6233ca2eb4f1dd..51361b75940560 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.cpp
@@ -100,14 +100,16 @@ GCNSubtarget &GCNSubtarget::initializeSubtargetDependencies(const Triple &TT,
   if (Gen == AMDGPUSubtarget::INVALID) {
     Gen = TT.getOS() == Triple::AMDHSA ? AMDGPUSubtarget::SEA_ISLANDS
                                        : AMDGPUSubtarget::SOUTHERN_ISLANDS;
-  }
-
-  if (!hasFeature(AMDGPU::FeatureWavefrontSize32) &&
-      !hasFeature(AMDGPU::FeatureWavefrontSize64)) {
+    // Assume wave64 for the unknown target, if not explicitly set.
+    if (getWavefrontSizeLog2() == 0)
+      WavefrontSizeLog2 = 6;
+  } else if (!hasFeature(AMDGPU::FeatureWavefrontSize32) &&
+             !hasFeature(AMDGPU::FeatureWavefrontSize64)) {
     // If there is no default wave size it must be a generation before gfx10,
     // these have FeatureWavefrontSize64 in their definition already. For gfx10+
     // set wave32 as a default.
     ToggleFeature(AMDGPU::FeatureWavefrontSize32);
+    WavefrontSizeLog2 = getGeneration() >= AMDGPUSubtarget::GFX10 ? 5 : 6;
   }
 
   // We don't support FP64 for EG/NI atm.
@@ -147,10 +149,6 @@ GCNSubtarget &GCNSubtarget::initializeSubtargetDependencies(const Triple &TT,
       !getFeatureBits().test(AMDGPU::FeatureCuMode))
     LocalMemorySize *= 2;
 
-  // Don't crash on invalid devices.
-  if (WavefrontSizeLog2 == 0)
-    WavefrontSizeLog2 = 5;
-
   HasFminFmaxLegacy = getGeneration() < AMDGPUSubtarget::VOLCANIC_ISLANDS;
   HasSMulHi = getGeneration() >= AMDGPUSubtarget::GFX9;
 
@@ -166,7 +164,7 @@ GCNSubtarget &GCNSubtarget::initializeSubtargetDependencies(const Triple &TT,
 
 void GCNSubtarget::checkSubtargetFeatures(const Function &F) const {
   LLVMContext &Ctx = F.getContext();
-  if (hasFeature(AMDGPU::FeatureWavefrontSize32) ==
+  if (hasFeature(AMDGPU::FeatureWavefrontSize32) &&
       hasFeature(AMDGPU::FeatureWavefrontSize64)) {
     Ctx.diagnose(DiagnosticInfoUnsupported(
         F, "must specify exactly one of wavefrontsize32 and wavefrontsize64"));
diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index f3f96940c1f44b..5eada4f003ece7 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -1564,6 +1564,14 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
     return getWavefrontSize() == 64;
   }
 
+  /// Returns if the wavesize of this subtarget is known reliable. This is false
+  /// only for the a default target-cpu that does not have an explicit
+  /// +wavefrontsize target feature.
+  bool isWaveSizeKnown() const {
+    return hasFeature(AMDGPU::FeatureWavefrontSize32) ||
+           hasFeature(AMDGPU::FeatureWavefrontSize64);
+  }
+
   const TargetRegisterClass *getBoolRC() const {
     return getRegisterInfo()->getBoolRC();
   }
diff --git a/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUInstPrinter.cpp b/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUInstPrinter.cpp
index 344028c4b48689..e21aa70c9859a0 100644
--- a/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUInstPrinter.cpp
+++ b/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUInstPrinter.cpp
@@ -649,9 +649,9 @@ void AMDGPUInstPrinter::printDefaultVccOperand(bool FirstOperand,
                                                raw_ostream &O) {
   if (!FirstOperand)
     O << ", ";
-  printRegOperand(STI.hasFeature(AMDGPU::FeatureWavefrontSize64)
-                      ? AMDGPU::VCC
-                      : AMDGPU::VCC_LO,
+  printRegOperand(STI.hasFeature(AMDGPU::FeatureWavefrontSize32)
+                      ? AMDGPU::VCC_LO
+                      : AMDGPU::VCC,
                   O, MRI);
   if (FirstOperand)
     O << ", ";