34 lines · plain
1; RUN: opt -data-layout=A5 -S -mtriple=amdgcn-unknown-unknown -passes=amdgpu-promote-alloca < %s | FileCheck -check-prefix=IR %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=ASM %s3 4 5@global_array0 = internal unnamed_addr addrspace(3) global [750 x [10 x i32]] poison, align 46@global_array1 = internal unnamed_addr addrspace(3) global [750 x [10 x i32]] poison, align 47 8; IR-LABEL: define amdgpu_kernel void @promote_alloca_size_256(ptr addrspace(1) captures(none) %out, ptr addrspace(1) captures(none) %in) {9; IR-NOT: alloca [10 x i32]10; ASM-LABEL: {{^}}promote_alloca_size_256:11; ASM: .amdgpu_lds llvm.amdgcn.kernel.promote_alloca_size_256.lds, 60000, 1612; ASM-NOT: .amdgpu_lds13 14define amdgpu_kernel void @promote_alloca_size_256(ptr addrspace(1) captures(none) %out, ptr addrspace(1) captures(none) %in) {15entry:16 %stack = alloca [10 x i32], align 4, addrspace(5)17 %tmp = load i32, ptr addrspace(1) %in, align 418 %arrayidx1 = getelementptr inbounds [10 x i32], ptr addrspace(5) %stack, i32 0, i32 %tmp19 store i32 4, ptr addrspace(5) %arrayidx1, align 420 %arrayidx2 = getelementptr inbounds i32, ptr addrspace(1) %in, i32 121 %tmp1 = load i32, ptr addrspace(1) %arrayidx2, align 422 %arrayidx3 = getelementptr inbounds [10 x i32], ptr addrspace(5) %stack, i32 0, i32 %tmp123 store i32 5, ptr addrspace(5) %arrayidx3, align 424 %tmp2 = load i32, ptr addrspace(5) %stack, align 425 store i32 %tmp2, ptr addrspace(1) %out, align 426 %arrayidx12 = getelementptr inbounds [10 x i32], ptr addrspace(5) %stack, i32 0, i32 127 %tmp3 = load i32, ptr addrspace(5) %arrayidx1228 %arrayidx13 = getelementptr inbounds i32, ptr addrspace(1) %out, i32 129 store i32 %tmp3, ptr addrspace(1) %arrayidx1330 store i32 %tmp3, ptr addrspace(3) @global_array031 store i32 %tmp3, ptr addrspace(3) @global_array132 ret void33}34