[llvm] AMDGPU: Fix asserting in DAG kernel argument lowering on v6i32 (PR #100528)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/100528

Remove this pointless assertion for the number of vector elements.

>From e0ad863e31d471d6dce796e7208e13d25ef77700 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Wed, 3 Jul 2024 12:04:04 +0200
Subject: [PATCH] AMDGPU: Fix asserting in DAG kernel argument lowering on
 v6i32

Remove this pointless assertion for the number of vector elements.
---
 llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp |  4 --
 .../AMDGPU/kernel-argument-dag-lowering.ll    | 42 +++++++++++++++++++
 2 files changed, 42 insertions(+), 4 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index ec386000302f1..2ad91de566323 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -1245,10 +1245,6 @@ void AMDGPUTargetLowering::analyzeFormalArgumentsCompute(
 
       // Round up vec3/vec5 argument.
       if (MemVT.isVector() && !MemVT.isPow2VectorType()) {
-        assert(MemVT.getVectorNumElements() == 3 ||
-               MemVT.getVectorNumElements() == 5 ||
-               (MemVT.getVectorNumElements() >= 9 &&
-                MemVT.getVectorNumElements() <= 12));
         MemVT = MemVT.getPow2VectorType(State.getContext());
       } else if (!MemVT.isSimple() && !MemVT.isVector()) {
         MemVT = MemVT.getRoundIntegerType(State.getContext());
diff --git a/llvm/test/CodeGen/AMDGPU/kernel-argument-dag-lowering.ll b/llvm/test/CodeGen/AMDGPU/kernel-argument-dag-lowering.ll
index f74f9a8f2bdd8..1a0aa09e2e40a 100644
--- a/llvm/test/CodeGen/AMDGPU/kernel-argument-dag-lowering.ll
+++ b/llvm/test/CodeGen/AMDGPU/kernel-argument-dag-lowering.ll
@@ -27,6 +27,48 @@ entry:
   ret void
 }
 
+; FUNC-LABEL: {{^}}v5i8_arg:
+; GCN: s_load_dwordx2 s[0:1], s[8:9], 0x0
+define amdgpu_kernel void @v5i8_arg(<5 x i8> %in) nounwind {
+  store <5 x i8> %in, ptr addrspace(1) null
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v6i8_arg:
+; GCN: s_load_dwordx2 s[0:1], s[8:9], 0x0
+define amdgpu_kernel void @v6i8_arg(<6 x i8> %in) nounwind {
+  store <6 x i8> %in, ptr addrspace(1) null
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v5i16_arg:
+; GCN: s_load_dwordx4 s[0:3], s[8:9], 0x0
+define amdgpu_kernel void @v5i16_arg(<5 x i16> %in) nounwind {
+  store <5 x i16> %in, ptr addrspace(1) null
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v6i16_arg:
+; GCN-DAG: s_load_dwordx4 s[0:3], s[8:9], 0x0
+define amdgpu_kernel void @v6i16_arg(<6 x i16> %in) nounwind {
+  store <6 x i16> %in, ptr addrspace(1) null
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v5i32_arg:
+; GCN: s_load_dwordx8 s[0:7], s[8:9], 0x0
+define amdgpu_kernel void @v5i32_arg(<5 x i32> %in) nounwind {
+  store <5 x i32> %in, ptr addrspace(1) null
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v6i32_arg:
+; GCN: s_load_dwordx8 s[0:7], s[8:9], 0x0
+define amdgpu_kernel void @v6i32_arg(<6 x i32> %in) nounwind {
+  store <6 x i32> %in, ptr addrspace(1) null
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}i65_arg:
 ; HSA-VI: s_load_dwordx4 s{{\[[0-9]+:[0-9]+\]}}, s[8:9], 0x0