[AMDGPU] Modify Dyn Alloca test to account for Machine-Verifier bug

[easyonaadit]

Machine-Verifier crashes in kernel functions,
but fails gracefully in device functions.

This is due to the buffer resource descriptor selected
during G-ISEL, before the fallback path.
Device functions use $sgpr0_sgpr1_sgpr2_sgpr3.
while Kernel functions select $private_rsrc_reg
where machine-verifier complains:
$private_rsrc_reg is not a SReg_128 register.

Modifying test case to capture both behaviors, this is related to #120063

[llvmbot]

@llvm/pr-subscribers-llvm-globalisel

@llvm/pr-subscribers-backend-amdgpu

Author: Aaditya (easyonaadit)

Changes

Machine-Verifier crashes in kernel functions,
but fails gracefully in device functions.

This is due to the buffer resource descriptor selected
during G-ISEL, before the fallback path.
Device functions use $sgpr0_sgpr1_sgpr2_sgpr3.
while Kernel functions select $private_rsrc_reg
where machine-verifier complains:
$private_rsrc_reg is not a SReg_128 register.

Modifying test case to capture both behaviors, this is related to #120063

---

Full diff: https://github.com/llvm/llvm-project/pull/120393.diff

1 Files Affected:

• (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/dynamic-alloca-divergent.ll (+12-7)

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/dynamic-alloca-divergent.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/dynamic-alloca-divergent.ll index 5dae7885f6bfb1..e0f22bc32b63c8 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/dynamic-alloca-divergent.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/dynamic-alloca-divergent.ll @@ -1,25 +1,30 @@ -; RUN: not llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -global-isel-abort=2 -pass-remarks-missed="gisel.*" -verify-machineinstrs -o /dev/null 2>&1 %s | FileCheck -check-prefix=ERR %s +; RUN: not --crash llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -global-isel-abort=2 -pass-remarks-missed="gisel.*" -verify-machineinstrs -o /dev/null 2>&1 %s | FileCheck -check-prefix=ERR_KERNEL %s +; RUN: not llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -global-isel-abort=2 -pass-remarks-missed="gisel.*" -o /dev/null 2>&1 %s | FileCheck -check-prefix=ERR %s + +; ERR_KERNEL: remark: <unknown>:0:0: cannot select: %{{[0-9]+}}:sreg_32(p5) = G_DYN_STACKALLOC %{{[0-9]+}}:vgpr(s32), 1 (in function: kernel_dynamic_stackalloc_vgpr_align4) +; ERR_KERNEL-NEXT: warning: Instruction selection used fallback path for kernel_dynamic_stackalloc_vgpr_align4 +; ERR_KERNEL-NEXT: error: <unknown>:0:0: in function kernel_dynamic_stackalloc_vgpr_align4 void (ptr addrspace(1)): unsupported dynamic alloca ; ERR: remark: <unknown>:0:0: cannot select: %{{[0-9]+}}:sreg_32(p5) = G_DYN_STACKALLOC %{{[0-9]+}}:vgpr(s32), 1 (in function: kernel_dynamic_stackalloc_vgpr_align4) ; ERR-NEXT: warning: Instruction selection used fallback path for kernel_dynamic_stackalloc_vgpr_align4 ; ERR-NEXT: error: <unknown>:0:0: in function kernel_dynamic_stackalloc_vgpr_align4 void (ptr addrspace(1)): unsupported dynamic alloca -; ERR: remark: <unknown>:0:0: cannot select: %{{[0-9]+}}:sreg_32(p5) = G_DYN_STACKALLOC %{{[0-9]+}}:vgpr(s32), 1 (in function: func_dynamic_stackalloc_vgpr_align4) -; ERR-NEXT: warning: Instruction selection used fallback path for func_dynamic_stackalloc_vgpr_align4 -; ERR-NEXT: error: <unknown>:0:0: in function func_dynamic_stackalloc_vgpr_align4 void (i32): unsupported dynamic alloca - define amdgpu_kernel void @kernel_dynamic_stackalloc_vgpr_align4(ptr addrspace(1) %ptr) { %id = call i32 @llvm.amdgcn.workitem.id.x() %gep = getelementptr i32, ptr addrspace(1) %ptr, i32 %id %n = load i32, ptr addrspace(1) %gep %alloca = alloca i32, i32 %n, align 4, addrspace(5) - store volatile ptr addrspace(5) %alloca, ptr addrspace(1) undef + store volatile i32 5, ptr addrspace(5) %alloca ret void } +; ERR: remark: <unknown>:0:0: cannot select: %{{[0-9]+}}:sreg_32(p5) = G_DYN_STACKALLOC %{{[0-9]+}}:vgpr(s32), 1 (in function: func_dynamic_stackalloc_vgpr_align4) +; ERR-NEXT: warning: Instruction selection used fallback path for func_dynamic_stackalloc_vgpr_align4 +; ERR-NEXT: error: <unknown>:0:0: in function func_dynamic_stackalloc_vgpr_align4 void (i32): unsupported dynamic alloca + define void @func_dynamic_stackalloc_vgpr_align4(i32 %n) { %alloca = alloca i32, i32 %n, align 4, addrspace(5) - store volatile ptr addrspace(5) %alloca, ptr addrspace(1) undef + store volatile i32 10, ptr addrspace(5) %alloca ret void } 

[arsenm]

his is due to the buffer resource descriptor selected
during G-ISEL, before the fallback path.
Device functions use $sgpr0_sgpr1_sgpr2_sgpr3.
while Kernel functions select $private_rsrc_reg
where machine-verifier complains:
$private_rsrc_reg is not a SReg_128 register.

This is just a bug that should be fixed, not a behavior to consider

[easyonaadit]

This is just a bug that should be fixed, not a behavior to consider

I need to modify the run line of this test file before pre-committing other tests for dynamic allocas. #120063
Another option is to remove the -verify-machineinstrs flag from the run line until this bug is resolved.

[arsenm]

This is just a bug that should be fixed, not a behavior to consider

I need to modify the run line of this test file before pre-committing other tests for dynamic allocas. #120063 Another option is to remove the -verify-machineinstrs flag from the run line until this bug is resolved.

You would need an explicit -verify-machineinstrs=0 to avoid breaking EXPENSIVE_CHECKS builds

[easyonaadit]

You would need an explicit -verify-machineinstrs=0 to avoid breaking EXPENSIVE_CHECKS builds

yess have done that instead.

[arsenm]

we ought to fix the machine verifier issue on fallback

[easyonaadit]

we ought to fix the machine verifier issue on fallback

Right, I'll be able to look into it after the dynamic alloca patch lands.