[llvm] [AMDGPU] Relocated getMaxNumWorkGroups function out of AMDGPUSubtarget (NFC) (PR #205636)

[Nikhil Kotikalapudi via llvm-commits]

https://github.com/nkotikal updated https://github.com/llvm/llvm-project/pull/205636

>From 419c3834e260411be8622f4ca2506b2033d71b5d Mon Sep 17 00:00:00 2001
From: Nikhil Kotikalapudi <Nikhil.Kotikalapudi at amd.com>
Date: Wed, 24 Jun 2026 15:42:13 -0500
Subject: [PATCH 1/2] [AMDGPU][NFC] Relocated getMaxNumWorkGroups function out
 of AMDGPUSubtarget and refactored

---
 llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp          | 3 +--
 llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp           | 7 -------
 llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h             | 3 ---
 llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp | 2 +-
 llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp     | 2 +-
 llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp      | 5 +++++
 llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h        | 3 +++
 7 files changed, 11 insertions(+), 14 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
index 0ddbb92783c39..7ebcf2ca47cdc 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
@@ -200,8 +200,7 @@ class AMDGPUInformationCache : public InformationCache {
   }
 
   SmallVector<unsigned> getMaxNumWorkGroups(const Function &F) {
-    const GCNSubtarget &ST = TM.getSubtarget<GCNSubtarget>(F);
-    return ST.getMaxNumWorkGroups(F);
+    return AMDGPU::getMaxNumWorkGroups(F);
   }
 
   /// Get code object version.
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
index fb4728609c877..3122eb60e160c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
@@ -430,10 +430,3 @@ const AMDGPUSubtarget &AMDGPUSubtarget::get(const TargetMachine &TM, const Funct
   return static_cast<const AMDGPUSubtarget &>(
       TM.getSubtarget<R600Subtarget>(F));
 }
-
-// FIXME: This has no reason to be in subtarget
-SmallVector<unsigned>
-AMDGPUSubtarget::getMaxNumWorkGroups(const Function &F) const {
-  return AMDGPU::getIntegerVecAttribute(F, "amdgpu-max-num-workgroups", 3,
-                                        std::numeric_limits<uint32_t>::max());
-}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
index 07746c087904d..f1ff6d80a55b3 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -294,9 +294,6 @@ class AMDGPUSubtarget {
   /// 2) dimension.
   unsigned getMaxWorkitemID(const Function &Kernel, unsigned Dimension) const;
 
-  /// Return the number of work groups for the function.
-  SmallVector<unsigned> getMaxNumWorkGroups(const Function &F) const;
-
   /// Return true if only a single workitem can be active in a wave.
   bool isSingleLaneExecution(const Function &Kernel) const;
 
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
index fe66a1a5d7242..4282d292b0c8d 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
@@ -1781,7 +1781,7 @@ bool GCNTTIImpl::shouldPrefetchAddressSpace(unsigned AS) const {
 void GCNTTIImpl::collectKernelLaunchBounds(
     const Function &F,
     SmallVectorImpl<std::pair<StringRef, int64_t>> &LB) const {
-  SmallVector<unsigned> MaxNumWorkgroups = ST->getMaxNumWorkGroups(F);
+  SmallVector<unsigned> MaxNumWorkgroups = AMDGPU::getMaxNumWorkGroups(F);
   LB.push_back({"amdgpu-max-num-workgroups[0]", MaxNumWorkgroups[0]});
   LB.push_back({"amdgpu-max-num-workgroups[1]", MaxNumWorkgroups[1]});
   LB.push_back({"amdgpu-max-num-workgroups[2]", MaxNumWorkgroups[2]});
diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
index 4be4ce28e6de5..59971923e4a5d 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
@@ -60,7 +60,7 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const Function &F,
   const GCNSubtarget &ST = *STI;
   FlatWorkGroupSizes = ST.getFlatWorkGroupSizes(F);
   WavesPerEU = ST.getWavesPerEU(F);
-  MaxNumWorkGroups = ST.getMaxNumWorkGroups(F);
+  MaxNumWorkGroups = AMDGPU::getMaxNumWorkGroups(F);
   assert(MaxNumWorkGroups.size() == 3);
 
   // Temporarily check both the attribute and the subtarget feature, until the
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
index 96571dd028b14..34779acd2b8d2 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
@@ -1756,6 +1756,11 @@ getIntegerVecAttribute(const Function &F, StringRef Name, unsigned Size) {
   return Vals;
 }
 
+SmallVector<unsigned> getMaxNumWorkGroups(const Function &F) {
+  return getIntegerVecAttribute(F, "amdgpu-max-num-workgroups", 3,
+                                std::numeric_limits<uint32_t>::max());
+}
+
 bool hasValueInRangeLikeMetadata(const MDNode &MD, int64_t Val) {
   assert((MD.getNumOperands() % 2 == 0) && "invalid number of operands!");
   for (unsigned I = 0, E = MD.getNumOperands() / 2; I != E; ++I) {
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
index 1623dc72d2810..3c42fb1a31005 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
@@ -1029,6 +1029,9 @@ SmallVector<unsigned> getIntegerVecAttribute(const Function &F, StringRef Name,
 std::optional<SmallVector<unsigned>>
 getIntegerVecAttribute(const Function &F, StringRef Name, unsigned Size);
 
+/// \returns The maximum number of workgroups for the function.
+SmallVector<unsigned> getMaxNumWorkGroups(const Function &F);
+
 /// Checks if \p Val is inside \p MD, a !range-like metadata.
 bool hasValueInRangeLikeMetadata(const MDNode &MD, int64_t Val);
 

>From 86f338cc42e24a0bbccbc99eb3c33f0a83b28f9c Mon Sep 17 00:00:00 2001
From: Nikhil Kotikalapudi <Nikhil.Kotikalapudi at amd.com>
Date: Sat, 27 Jun 2026 23:17:09 -0500
Subject: [PATCH 2/2] removed unnecessary wrapper

---
 llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
index 7ebcf2ca47cdc..3ce0d5b8b800a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
@@ -199,10 +199,6 @@ class AMDGPUInformationCache : public InformationCache {
     return {ST.getMinFlatWorkGroupSize(), ST.getMaxFlatWorkGroupSize()};
   }
 
-  SmallVector<unsigned> getMaxNumWorkGroups(const Function &F) {
-    return AMDGPU::getMaxNumWorkGroups(F);
-  }
-
   /// Get code object version.
   unsigned getCodeObjectVersion() const { return CodeObjectVersion; }
 
@@ -1030,7 +1026,7 @@ struct AAAMDMaxNumWorkgroups
     Function *F = getAssociatedFunction();
     auto &InfoCache = static_cast<AMDGPUInformationCache &>(A.getInfoCache());
 
-    SmallVector<unsigned> MaxNumWorkgroups = InfoCache.getMaxNumWorkGroups(*F);
+    SmallVector<unsigned> MaxNumWorkgroups = AMDGPU::getMaxNumWorkGroups(*F);
 
     X.takeKnownMinimum(MaxNumWorkgroups[0]);
     Y.takeKnownMinimum(MaxNumWorkgroups[1]);