[llvm] [AMDGPU] Move kernarg preload logic to separate pass (PR #130434)

[Christudasan Devadasan via llvm-commits]

================
@@ -0,0 +1,358 @@
+//===- AMDGPUPreloadKernelArguments.cpp - Preload Kernel Arguments --------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file This pass preloads kernel arguments into user_data SGPRs before kernel
+/// execution begins. The number of registers available for preloading depends
+/// on the number of free user SGPRs, up to the hardware's maximum limit.
+/// Implicit arguments enabled in the kernel descriptor are allocated first,
+/// followed by SGPRs used for preloaded kernel arguments. (Reference:
+/// https://llvm.org/docs/AMDGPUUsage.html#initial-kernel-execution-state)
+/// Additionally, hidden kernel arguments may be preloaded, in which case they
+/// are appended to the kernel signature after explicit arguments. Preloaded
+/// arguments will be marked with `inreg`.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "AMDGPUTargetMachine.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicsAMDGPU.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/PassManager.h"
+#include "llvm/IR/Verifier.h"
+#include "llvm/Pass.h"
+
+#define DEBUG_TYPE "amdgpu-preload-kernel-arguments"
+
+using namespace llvm;
+
+static cl::opt<unsigned> KernargPreloadCount(
+    "amdgpu-kernarg-preload-count",
+    cl::desc("How many kernel arguments to preload onto SGPRs"), cl::init(0));
+
+namespace {
+
+class AMDGPUPreloadKernelArgumentsLegacy : public ModulePass {
+  const AMDGPUTargetMachine *TM;
+
+public:
+  static char ID;
+  explicit AMDGPUPreloadKernelArgumentsLegacy(
+      const AMDGPUTargetMachine *TM = nullptr);
+
+  StringRef getPassName() const override {
+    return "AMDGPU Preload Kernel Arguments";
+  }
+
+  bool runOnModule(Module &M) override;
+};
+
+class PreloadKernelArgInfo {
+private:
+  Function &F;
+  const GCNSubtarget &ST;
+  unsigned NumFreeUserSGPRs;
+
+  enum HiddenArg : unsigned {
+    HIDDEN_BLOCK_COUNT_X,
+    HIDDEN_BLOCK_COUNT_Y,
+    HIDDEN_BLOCK_COUNT_Z,
+    HIDDEN_GROUP_SIZE_X,
+    HIDDEN_GROUP_SIZE_Y,
+    HIDDEN_GROUP_SIZE_Z,
+    HIDDEN_REMAINDER_X,
+    HIDDEN_REMAINDER_Y,
+    HIDDEN_REMAINDER_Z,
+    END_HIDDEN_ARGS
+  };
+
+  // Stores information about a specific hidden argument.
+  struct HiddenArgInfo {
+    // Offset in bytes from the location in the kernearg segment pointed to by
+    // the implicitarg pointer.
+    uint8_t Offset;
+    // The size of the hidden argument in bytes.
+    uint8_t Size;
+    // The name of the hidden argument in the kernel signature.
+    const char *Name;
+  };
+
+  static constexpr HiddenArgInfo HiddenArgs[END_HIDDEN_ARGS] = {
+      {0, 4, "_hidden_block_count_x"}, {4, 4, "_hidden_block_count_y"},
+      {8, 4, "_hidden_block_count_z"}, {12, 2, "_hidden_group_size_x"},
+      {14, 2, "_hidden_group_size_y"}, {16, 2, "_hidden_group_size_z"},
+      {18, 2, "_hidden_remainder_x"},  {20, 2, "_hidden_remainder_y"},
+      {22, 2, "_hidden_remainder_z"}};
+
+  static HiddenArg getHiddenArgFromOffset(unsigned Offset) {
+    for (unsigned I = 0; I < END_HIDDEN_ARGS; ++I)
+      if (HiddenArgs[I].Offset == Offset)
+        return static_cast<HiddenArg>(I);
+
+    return END_HIDDEN_ARGS;
+  }
+
+  static Type *getHiddenArgType(LLVMContext &Ctx, HiddenArg HA) {
+    if (HA < END_HIDDEN_ARGS)
+      return Type::getIntNTy(Ctx, HiddenArgs[HA].Size * 8);
+
+    llvm_unreachable("Unexpected hidden argument.");
+  }
+
+  static const char *getHiddenArgName(HiddenArg HA) {
+    if (HA < END_HIDDEN_ARGS) {
+      return HiddenArgs[HA].Name;
+    }
----------------
cdevadas wrote:

```suggestion
    if (HA < END_HIDDEN_ARGS)
      return HiddenArgs[HA].Name;
```

https://github.com/llvm/llvm-project/pull/130434