Document attribute, fix issues with pass being run multiple times.

Author: kerbowa

llvm/docs/AMDGPUUsage.rst

@@ -1639,6 +1639,10 @@ The AMDGPU backend supports the following LLVM IR attributes.
                                              function which requires AGPRs is reached through any function marked
                                              with this attribute.
 
+     "amdgpu-hidden-argument"                This attribute is used internally by the backend to mark function arguments
+                                             as hidden. Hidden arguments are managed by the compiler and are not part of
+                                             the explicit arguments supplied by the user.
+
      ======================================= ==========================================================
 
 Calling Conventions
@@ -5856,6 +5860,11 @@ may insert a trap instruction at the start of the kernel prologue to manage
 situations where kernarg preloading is attempted on hardware with incompatible
 firmware.
 
+With code object V5 and later, hidden kernel arguments that are normally accessed
+through the Implicit Argument Ptr, may be preloaded into User SGPRs. These
+arguments are added to the kernel function signature and are marked with the
+attribute "amdgpu-hidden-argument". (See :ref:`amdgpu-llvm-ir-attributes-table`).
+
 .. _amdgpu-amdhsa-kernel-prolog:
 
 Kernel Prolog

llvm/lib/Target/AMDGPU/AMDGPULowerKernelArguments.cpp

@@ -325,6 +325,12 @@ static bool lowerKernelArguments(Function &F, const TargetMachine &TM) {
     uint64_t LastExplicitArgOffset = ExplicitArgOffset;
     ExplicitArgOffset = alignTo(ExplicitArgOffset, ABITypeAlign) + AllocSize;
 
+    // Guard against the situation where hidden arguments have already been lowered
+    // and added to the kernel function signiture, i.e. in a situation where this
+    // pass has run twice.
+    if (Arg.hasAttribute("amdgpu-hidden-argument"))
+      break;
+
     // Try to preload this argument into user SGPRs.
     if (Arg.hasInRegAttr() && InPreloadSequence && ST.hasKernargPreload() &&
         !Arg.getType()->isAggregateType())

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

@@ -2546,7 +2546,9 @@ void SITargetLowering::allocatePreloadKernArgSGPRs(
         if (OffsetBefore != LastExplicitArgOffset) {
           unsigned PaddingSGPRs =
               alignTo(LastExplicitArgOffset - OffsetBefore, 4) / 4;
-          Info.allocateUserSGPRs(*Subtarget, PaddingSGPRs);
+          if (!Info.allocateUserSGPRs(*Subtarget, PaddingSGPRs))
+            break;
+
           ArgOffset += PaddingSGPRs * 4;
         }
         AlignedForImplictArgs = true;