Merge branch 'main' into users/meinersbur/flang_runtime_split-headersusers/meinersbur/flang_runtime_split-headers

1 files changed, 182 insertions, 2 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/dynamic_stackalloc.ll b/llvm/test/CodeGen/AMDGPU/dynamic_stackalloc.ll
index 1c093bf31ea7..73aa87e5c55d 100644
--- a/llvm/test/CodeGen/AMDGPU/dynamic_stackalloc.ll
+++ b/llvm/test/CodeGen/AMDGPU/dynamic_stackalloc.ll
@@ -5,8 +5,188 @@ target datalayout = "A5"
 
 ; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
 
-define amdgpu_kernel void @test_dynamic_stackalloc(ptr addrspace(1) %out, i32 %n) {
+define amdgpu_kernel void @test_dynamic_stackalloc_kernel_uniform(i32 %n) {
   %alloca = alloca i32, i32 %n, addrspace(5)
-  store volatile i32 0, ptr addrspace(5) %alloca
+  store volatile i32 123, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define amdgpu_kernel void @test_dynamic_stackalloc_kernel_uniform_over_aligned(i32 %n) {
+  %alloca = alloca i32, i32 %n, align 128, addrspace(5)
+  store volatile i32 10, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define amdgpu_kernel void @test_dynamic_stackalloc_kernel_uniform_under_aligned(i32 %n) {
+  %alloca = alloca i32, i32 %n, align 2, addrspace(5)
+  store volatile i32 22, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define amdgpu_kernel void @test_dynamic_stackalloc_kernel_divergent() {
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca = alloca float, i32 %idx, addrspace(5)
+  store volatile i32 123, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define amdgpu_kernel void @test_dynamic_stackalloc_kernel_divergent_over_aligned() {
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca = alloca i32, i32 %idx, align 128, addrspace(5)
+  store volatile i32 444, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define amdgpu_kernel void @test_dynamic_stackalloc_kernel_divergent_under_aligned() {
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca = alloca i128, i32 %idx, align 2, addrspace(5)
+  store volatile i32 666, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define amdgpu_kernel void @test_dynamic_stackalloc_kernel_multiple_allocas(i32 %n, i32 %m) {
+entry:
+  %cond = icmp eq i32 %n, 0
+  %alloca1 = alloca i32, i32 8, addrspace(5)
+  %alloca2 = alloca i17, i32 %n, addrspace(5)
+  br i1 %cond, label %bb.0, label %bb.1
+bb.0:
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca3 = alloca i32, i32 %m, align 64, addrspace(5)
+  %alloca4 = alloca i32, i32 %idx, align 4, addrspace(5)
+  store volatile i32 3, ptr addrspace(5) %alloca3
+  store volatile i32 4, ptr addrspace(5) %alloca4
+  br label %bb.1
+bb.1:
+  store volatile i32 1, ptr addrspace(5) %alloca1
+  store volatile i32 2, ptr addrspace(5) %alloca2
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define amdgpu_kernel void @test_dynamic_stackalloc_kernel_control_flow(i32 %n, i32 %m) {
+entry:
+  %cond = icmp eq i32 %n, 0
+  br i1 %cond, label %bb.0, label %bb.1
+bb.0:
+  %alloca2 = alloca i32, i32 %m, align 64, addrspace(5)
+  store volatile i32 2, ptr addrspace(5) %alloca2
+  br label %bb.2
+bb.1:
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca1 = alloca i32, i32 %idx, align 4, addrspace(5)
+  store volatile i32 1, ptr addrspace(5) %alloca1
+  br label %bb.2
+bb.2:
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define void @test_dynamic_stackalloc_device_uniform(i32 %n) {
+  %alloca = alloca i32, i32 %n, addrspace(5)
+  store volatile i32 123, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define void @test_dynamic_stackalloc_device_uniform_over_aligned(i32 %n) {
+  %alloca = alloca i32, i32 %n, align 128, addrspace(5)
+  store volatile i32 10, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define void @test_dynamic_stackalloc_device_uniform_under_aligned(i32 %n) {
+  %alloca = alloca i32, i32 %n, align 2, addrspace(5)
+  store volatile i32 22, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define void @test_dynamic_stackalloc_device_divergent() {
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca = alloca i32, i32 %idx, addrspace(5)
+  store volatile i32 123, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define void @test_dynamic_stackalloc_device_divergent_over_aligned() {
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca = alloca i32, i32 %idx, align 128, addrspace(5)
+  store volatile i32 444, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define void @test_dynamic_stackalloc_device_divergent_under_aligned() {
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca = alloca i32, i32 %idx, align 2, addrspace(5)
+  store volatile i32 666, ptr addrspace(5) %alloca
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define void @test_dynamic_stackalloc_device_multiple_allocas(i32 %n, i32 %m) {
+entry:
+  %cond = icmp eq i32 %n, 0
+  %alloca1 = alloca i32, i32 8, addrspace(5)
+  %alloca2 = alloca i32, i32 %n, addrspace(5)
+  br i1 %cond, label %bb.0, label %bb.1
+bb.0:
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca3 = alloca i32, i32 %m, align 64, addrspace(5)
+  %alloca4 = alloca i32, i32 %idx, align 4, addrspace(5)
+  store volatile i32 3, ptr addrspace(5) %alloca3
+  store volatile i32 4, ptr addrspace(5) %alloca4
+  br label %bb.1
+bb.1:
+  store volatile i32 1, ptr addrspace(5) %alloca1
+  store volatile i32 2, ptr addrspace(5) %alloca2
+  ret void
+}
+
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+; CHECK: in function test_dynamic_stackalloc{{.*}}: unsupported dynamic alloca
+
+define void @test_dynamic_stackalloc_device_control_flow(i32 %n, i32 %m) {
+entry:
+  %cond = icmp eq i32 %n, 0
+  br i1 %cond, label %bb.0, label %bb.1
+bb.0:
+  %idx = call i32 @llvm.amdgcn.workitem.id.x()
+  %alloca1 = alloca i32, i32 %idx, align 4, addrspace(5)
+  store volatile i32 1, ptr addrspace(5) %alloca1
+  br label %bb.2
+bb.1:
+  %alloca2 = alloca i32, i32 %m, align 64, addrspace(5)
+  store volatile i32 2, ptr addrspace(5) %alloca2
+  br label %bb.2
+bb.2:
   ret void
 }