[llvm] [AMDGPU] Add dynamic LDS size implicit kernel argument to CO-v5 (PR #65273)

[via llvm-commits]

https://github.com/skc7 updated https://github.com/llvm/llvm-project/pull/65273

>From 6a34b8e558ea741c368d275d4a7621a7bfd8d99b Mon Sep 17 00:00:00 2001
From: skc7 <Krishna.Sankisa at amd.com>
Date: Mon, 4 Sep 2023 20:16:51 +0530
Subject: [PATCH] [AMDGPU] Add dynamic LDS size implicit kernel argument to
 CO-v5

hidden_dynamic_lds_size argument will be added in the reserved
section at offset 120 of the implicit argument layout.
---
 llvm/docs/AMDGPUUsage.rst                               | 3 +++
 llvm/lib/BinaryFormat/AMDGPUMetadataVerifier.cpp        | 1 +
 llvm/lib/Target/AMDGPU/AMDGPUHSAMetadataStreamer.cpp    | 5 ++++-
 llvm/test/CodeGen/AMDGPU/hsa-metadata-hidden-args-v5.ll | 5 ++++-
 4 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/llvm/docs/AMDGPUUsage.rst b/llvm/docs/AMDGPUUsage.rst
index 8022816d7e616d3..a48f25d6c8070b1 100644
--- a/llvm/docs/AMDGPUUsage.rst
+++ b/llvm/docs/AMDGPUUsage.rst
@@ -4024,6 +4024,9 @@ Code object V5 metadata is the same as
                                                        buffer that conforms to the requirements of the malloc/free
                                                        device library V1 version implementation.
 
+                                                     "hidden_dynamic_lds_size"
+                                                       Size of the dynamically allocated LDS memory is passed in the kernarg.
+
                                                      "hidden_private_base"
                                                        The high 32 bits of the flat addressing private aperture base.
                                                        Only used by GFX8 to allow conversion between private segment
diff --git a/llvm/lib/BinaryFormat/AMDGPUMetadataVerifier.cpp b/llvm/lib/BinaryFormat/AMDGPUMetadataVerifier.cpp
index 35a79ec04b6e767..f94940eecae20d9 100644
--- a/llvm/lib/BinaryFormat/AMDGPUMetadataVerifier.cpp
+++ b/llvm/lib/BinaryFormat/AMDGPUMetadataVerifier.cpp
@@ -135,6 +135,7 @@ bool MetadataVerifier::verifyKernelArgs(msgpack::DocNode &Node) {
                                .Case("hidden_default_queue", true)
                                .Case("hidden_completion_action", true)
                                .Case("hidden_multigrid_sync_arg", true)
+                               .Case("hidden_dynamic_lds_size", true)
                                .Case("hidden_private_base", true)
                                .Case("hidden_shared_base", true)
                                .Case("hidden_queue_ptr", true)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUHSAMetadataStreamer.cpp b/llvm/lib/Target/AMDGPU/AMDGPUHSAMetadataStreamer.cpp
index 5060cd3aec581ce..fe27774776d450e 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUHSAMetadataStreamer.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUHSAMetadataStreamer.cpp
@@ -663,7 +663,10 @@ void MetadataStreamerMsgPackV5::emitHiddenKernelArgs(
     Offset += 8; // Skipped.
   }
 
-  Offset += 72; // Reserved.
+  // emit argument for hidden dynamic lds size
+  emitKernelArg(DL, Int32Ty, Align(4), "hidden_dynamic_lds_size", Offset, Args);
+
+  Offset += 68; // Reserved.
 
   // hidden_private_base and hidden_shared_base are only when the subtarget has
   // ApertureRegs.
diff --git a/llvm/test/CodeGen/AMDGPU/hsa-metadata-hidden-args-v5.ll b/llvm/test/CodeGen/AMDGPU/hsa-metadata-hidden-args-v5.ll
index cb30d668674c316..690a0b02a5ff754 100644
--- a/llvm/test/CodeGen/AMDGPU/hsa-metadata-hidden-args-v5.ll
+++ b/llvm/test/CodeGen/AMDGPU/hsa-metadata-hidden-args-v5.ll
@@ -81,13 +81,16 @@
 ; CHECK-NEXT:      - .offset:         136
 ; CHECK-NEXT:        .size:           8
 ; CHECK-NEXT:        .value_kind:     hidden_completion_action
+; CHECK:          - .offset:          144
+; CHECK-NEXT:        .size:           4
+; CHECK-NEXT:        .value_kind:     hidden_dynamic_lds_size
 ; GFX8-NEXT:      - .offset:         216
 ; GFX8-NEXT:        .size:           4
 ; GFX8-NEXT:        .value_kind:     hidden_private_base
 ; GFX8-NEXT:      - .offset:         220
 ; GFX8-NEXT:        .size:           4
 ; GFX8-NEXT:        .value_kind:     hidden_shared_base
-; CHECK:      - .offset:         224
+; CHECK:          - .offset:          224
 ; CHECK-NEXT:        .size:           8
 ; CHECK-NEXT:        .value_kind:     hidden_queue_ptr