InferAddressSpaces: Add more baseline tests for assume handling

[arsenm]

No description provided.

[arsenm]

• InferAddressSpaces: Add more baseline tests for assume handling #167611 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Matt Arsenault (arsenm)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/167611.diff

1 Files Affected:

• (modified) llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll (+200-8)

diff --git a/llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll b/llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll index e0c80c0389541..32dca860a7ded 100644 --- a/llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll +++ b/llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll @@ -1,8 +1,8 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=infer-address-spaces -o - %s | FileCheck %s -define float @f0(ptr %p) { -; CHECK-LABEL: define float @f0( +define float @assume_is_shared_gep(ptr %p) { +; CHECK-LABEL: define float @assume_is_shared_gep( ; CHECK-SAME: ptr [[P:%.*]]) { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[P]]) @@ -24,8 +24,8 @@ entry: ret float %load } -define float @f1(ptr %p) { -; CHECK-LABEL: define float @f1( +define float @assume_is_private_gep(ptr %p) { +; CHECK-LABEL: define float @assume_is_private_gep( ; CHECK-SAME: ptr [[P:%.*]]) { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[P]]) @@ -47,8 +47,8 @@ entry: ret float %load } -define float @f2(ptr %p) { -; CHECK-LABEL: define float @f2( +define float @assume_not_private_and_not_shared_gep(ptr %p) { +; CHECK-LABEL: define float @assume_not_private_and_not_shared_gep( ; CHECK-SAME: ptr [[P:%.*]]) { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[P]]) @@ -78,8 +78,8 @@ entry: ret float %load } -define float @g0(i32 %c, ptr %p) { -; CHECK-LABEL: define float @g0( +define float @conditionally_assume_is_shared_gep(i32 %c, ptr %p) { +; CHECK-LABEL: define float @conditionally_assume_is_shared_gep( ; CHECK-SAME: i32 [[C:%.*]], ptr [[P:%.*]]) { ; CHECK-NEXT: [[ENTRY:.*]]: ; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[C]], 0 @@ -127,6 +127,198 @@ if.end: ret float %add2 } +define float @conditionally_assume_is_shared_else_assume_private(i32 %c, ptr %p) { +; CHECK-LABEL: define float @conditionally_assume_is_shared_else_assume_private( +; CHECK-SAME: i32 [[C:%.*]], ptr [[P:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[C]], 0 +; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label %[[IF_THEN_SHARED:.*]], label %[[IF_THEN_PRIVATE:.*]] +; CHECK: [[IF_THEN_SHARED]]: +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[P]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_SHARED]]) +; CHECK-NEXT: [[WORKITEM_ID_X_0:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[WORKITEM_ID_X_0]] to i64 +; CHECK-NEXT: [[TMP0:%.*]] = addrspacecast ptr [[P]] to ptr addrspace(3) +; CHECK-NEXT: [[ARRAYIDX0:%.*]] = getelementptr inbounds float, ptr addrspace(3) [[TMP0]], i64 [[IDXPROM]] +; CHECK-NEXT: [[LOAD0:%.*]] = load float, ptr addrspace(3) [[ARRAYIDX0]], align 4 +; CHECK-NEXT: [[ADD0:%.*]] = fadd float [[LOAD0]], 4.000000e+00 +; CHECK-NEXT: br label %[[IF_END:.*]] +; CHECK: [[IF_THEN_PRIVATE]]: +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[P]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_PRIVATE]]) +; CHECK-NEXT: [[WORKITEM_ID_X_1:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[IDXPROM1:%.*]] = zext i32 [[WORKITEM_ID_X_1]] to i64 +; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[P]] to ptr addrspace(5) +; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds float, ptr addrspace(5) [[TMP1]], i64 [[IDXPROM1]] +; CHECK-NEXT: [[LOAD1:%.*]] = load float, ptr addrspace(5) [[ARRAYIDX1]], align 4 +; CHECK-NEXT: [[ADD1:%.*]] = fadd float [[LOAD1]], 4.000000e+00 +; CHECK-NEXT: br label %[[IF_END]] +; CHECK: [[IF_END]]: +; CHECK-NEXT: [[PHI:%.*]] = phi float [ [[ADD0]], %[[IF_THEN_SHARED]] ], [ [[ADD1]], %[[IF_THEN_PRIVATE]] ] +; CHECK-NEXT: ret float [[PHI]] +; +entry: + %tobool.not = icmp eq i32 %c, 0 + br i1 %tobool.not, label %if.then.shared, label %if.then.private + +if.then.shared: + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %p) + tail call void @llvm.assume(i1 %is.shared) + %workitem.id.x.0 = tail call i32 @llvm.amdgcn.workitem.id.x() + %idxprom = zext i32 %workitem.id.x.0 to i64 + %arrayidx0 = getelementptr inbounds float, ptr %p, i64 %idxprom + %load0 = load float, ptr %arrayidx0, align 4 + %add0 = fadd float %load0, 4.0 + br label %if.end + +if.then.private: + %is.private = call i1 @llvm.amdgcn.is.private(ptr %p) + tail call void @llvm.assume(i1 %is.private) + %workitem.id.x.1 = tail call i32 @llvm.amdgcn.workitem.id.x() + %idxprom1 = zext i32 %workitem.id.x.1 to i64 + %arrayidx1 = getelementptr inbounds float, ptr %p, i64 %idxprom1 + %load1 = load float, ptr %arrayidx1, align 4 + %add1 = fadd float %load1, 4.0 + br label %if.end + +if.end: + %phi = phi float [ %add0, %if.then.shared ], [ %add1, %if.then.private ] + ret float %phi +} + +define float @assume_func_arg_is_shared_load(ptr %flat.ptr) { +; CHECK-LABEL: define float @assume_func_arg_is_shared_load( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) { +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[FLAT_PTR]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_SHARED]]) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr [[FLAT_PTR]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %flat.ptr) + tail call void @llvm.assume(i1 %is.shared) + %load = load float, ptr %flat.ptr, align 4 + ret float %load +} + +define float @assume_func_arg_is_private_load(ptr %flat.ptr) { +; CHECK-LABEL: define float @assume_func_arg_is_private_load( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) { +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[FLAT_PTR]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_PRIVATE]]) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr [[FLAT_PTR]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.private = call i1 @llvm.amdgcn.is.private(ptr %flat.ptr) + tail call void @llvm.assume(i1 %is.private) + %load = load float, ptr %flat.ptr, align 4 + ret float %load +} + +define float @assume_func_arg_is_not_shared_not_private(ptr %flat.ptr) { +; CHECK-LABEL: define float @assume_func_arg_is_not_shared_not_private( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) { +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[FLAT_PTR]]) +; CHECK-NEXT: [[NOT_PRIVATE:%.*]] = xor i1 [[IS_PRIVATE]], true +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[FLAT_PTR]]) +; CHECK-NEXT: [[NOT_SHARED:%.*]] = xor i1 [[IS_SHARED]], true +; CHECK-NEXT: [[NOT_PRIVATE_AND_NOT_SHARED:%.*]] = and i1 [[NOT_PRIVATE]], [[NOT_SHARED]] +; CHECK-NEXT: tail call void @llvm.assume(i1 [[NOT_PRIVATE_AND_NOT_SHARED]]) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr [[FLAT_PTR]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.private = call i1 @llvm.amdgcn.is.private(ptr %flat.ptr) + %not.private = xor i1 %is.private, true + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %flat.ptr) + %not.shared = xor i1 %is.shared, true + %not.private.and.not.shared = and i1 %not.private, %not.shared + tail call void @llvm.assume(i1 %not.private.and.not.shared) + %load = load float, ptr %flat.ptr, align 4 + ret float %load +} + +define float @assume_func_arg_is_not_private_load(ptr %flat.ptr) { +; CHECK-LABEL: define float @assume_func_arg_is_not_private_load( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) { +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[FLAT_PTR]]) +; CHECK-NEXT: [[NOT_IS_PRIVATE:%.*]] = xor i1 [[IS_PRIVATE]], true +; CHECK-NEXT: tail call void @llvm.assume(i1 [[NOT_IS_PRIVATE]]) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr [[FLAT_PTR]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.private = call i1 @llvm.amdgcn.is.private(ptr %flat.ptr) + %not.is.private = xor i1 %is.private, true + tail call void @llvm.assume(i1 %not.is.private) + %load = load float, ptr %flat.ptr, align 4 + ret float %load +} + +define i64 @assume_func_arg_is_not_private_atomicrmw(ptr %flat.ptr, i64 %val) { +; CHECK-LABEL: define i64 @assume_func_arg_is_not_private_atomicrmw( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]], i64 [[VAL:%.*]]) { +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[FLAT_PTR]]) +; CHECK-NEXT: [[NOT_IS_PRIVATE:%.*]] = xor i1 [[IS_PRIVATE]], true +; CHECK-NEXT: tail call void @llvm.assume(i1 [[NOT_IS_PRIVATE]]) +; CHECK-NEXT: [[RMW:%.*]] = atomicrmw sub ptr [[FLAT_PTR]], i64 [[VAL]] seq_cst, align 4 +; CHECK-NEXT: ret i64 [[RMW]] +; + %is.private = call i1 @llvm.amdgcn.is.private(ptr %flat.ptr) + %not.is.private = xor i1 %is.private, true + tail call void @llvm.assume(i1 %not.is.private) + %rmw = atomicrmw sub ptr %flat.ptr, i64 %val seq_cst, align 4 + ret i64 %rmw +} + +define float @contradictory_assume_after_gep_same_block(ptr %p) { +; CHECK-LABEL: define float @contradictory_assume_after_gep_same_block( +; CHECK-SAME: ptr [[P:%.*]]) { +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[P]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_SHARED]]) +; CHECK-NEXT: [[WORKITEM_ID_X:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[WORKITEM_ID_X]] to i64 +; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[P]] to ptr addrspace(3) +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr addrspace(3) [[TMP1]], i64 [[IDXPROM]] +; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[P]], i64 [[IDXPROM]] +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[TMP2]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 false) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr addrspace(3) [[GEP]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %p) + tail call void @llvm.assume(i1 %is.shared) + %workitem.id.x = tail call i32 @llvm.amdgcn.workitem.id.x() + %idxprom = zext i32 %workitem.id.x to i64 + %gep = getelementptr inbounds float, ptr %p, i64 %idxprom + %is.private = call i1 @llvm.amdgcn.is.private(ptr %gep) + tail call void @llvm.assume(i1 %is.private) + %load = load float, ptr %gep, align 4 + ret float %load +} + +define float @contradictory_assume_argument_same_block(ptr %p) { +; CHECK-LABEL: define float @contradictory_assume_argument_same_block( +; CHECK-SAME: ptr [[P:%.*]]) { +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[P]]) +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[P]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_SHARED]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_PRIVATE]]) +; CHECK-NEXT: [[WORKITEM_ID_X:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[WORKITEM_ID_X]] to i64 +; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[P]] to ptr addrspace(3) +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr addrspace(3) [[TMP1]], i64 [[IDXPROM]] +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr addrspace(3) [[GEP]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %p) + %is.private = call i1 @llvm.amdgcn.is.private(ptr %p) + tail call void @llvm.assume(i1 %is.shared) + tail call void @llvm.assume(i1 %is.private) + %workitem.id.x = tail call i32 @llvm.amdgcn.workitem.id.x() + %idxprom = zext i32 %workitem.id.x to i64 + %gep = getelementptr inbounds float, ptr %p, i64 %idxprom + %load = load float, ptr %gep, align 4 + ret float %load +} + declare void @llvm.assume(i1) declare i1 @llvm.amdgcn.is.shared(ptr nocapture) declare i1 @llvm.amdgcn.is.private(ptr nocapture) 

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/167611.diff

1 Files Affected:

• (modified) llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll (+200-8)

diff --git a/llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll b/llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll index e0c80c0389541..32dca860a7ded 100644 --- a/llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll +++ b/llvm/test/Transforms/InferAddressSpaces/AMDGPU/builtin-assumed-addrspace.ll @@ -1,8 +1,8 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=infer-address-spaces -o - %s | FileCheck %s -define float @f0(ptr %p) { -; CHECK-LABEL: define float @f0( +define float @assume_is_shared_gep(ptr %p) { +; CHECK-LABEL: define float @assume_is_shared_gep( ; CHECK-SAME: ptr [[P:%.*]]) { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[P]]) @@ -24,8 +24,8 @@ entry: ret float %load } -define float @f1(ptr %p) { -; CHECK-LABEL: define float @f1( +define float @assume_is_private_gep(ptr %p) { +; CHECK-LABEL: define float @assume_is_private_gep( ; CHECK-SAME: ptr [[P:%.*]]) { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[P]]) @@ -47,8 +47,8 @@ entry: ret float %load } -define float @f2(ptr %p) { -; CHECK-LABEL: define float @f2( +define float @assume_not_private_and_not_shared_gep(ptr %p) { +; CHECK-LABEL: define float @assume_not_private_and_not_shared_gep( ; CHECK-SAME: ptr [[P:%.*]]) { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[P]]) @@ -78,8 +78,8 @@ entry: ret float %load } -define float @g0(i32 %c, ptr %p) { -; CHECK-LABEL: define float @g0( +define float @conditionally_assume_is_shared_gep(i32 %c, ptr %p) { +; CHECK-LABEL: define float @conditionally_assume_is_shared_gep( ; CHECK-SAME: i32 [[C:%.*]], ptr [[P:%.*]]) { ; CHECK-NEXT: [[ENTRY:.*]]: ; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[C]], 0 @@ -127,6 +127,198 @@ if.end: ret float %add2 } +define float @conditionally_assume_is_shared_else_assume_private(i32 %c, ptr %p) { +; CHECK-LABEL: define float @conditionally_assume_is_shared_else_assume_private( +; CHECK-SAME: i32 [[C:%.*]], ptr [[P:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[C]], 0 +; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label %[[IF_THEN_SHARED:.*]], label %[[IF_THEN_PRIVATE:.*]] +; CHECK: [[IF_THEN_SHARED]]: +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[P]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_SHARED]]) +; CHECK-NEXT: [[WORKITEM_ID_X_0:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[WORKITEM_ID_X_0]] to i64 +; CHECK-NEXT: [[TMP0:%.*]] = addrspacecast ptr [[P]] to ptr addrspace(3) +; CHECK-NEXT: [[ARRAYIDX0:%.*]] = getelementptr inbounds float, ptr addrspace(3) [[TMP0]], i64 [[IDXPROM]] +; CHECK-NEXT: [[LOAD0:%.*]] = load float, ptr addrspace(3) [[ARRAYIDX0]], align 4 +; CHECK-NEXT: [[ADD0:%.*]] = fadd float [[LOAD0]], 4.000000e+00 +; CHECK-NEXT: br label %[[IF_END:.*]] +; CHECK: [[IF_THEN_PRIVATE]]: +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[P]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_PRIVATE]]) +; CHECK-NEXT: [[WORKITEM_ID_X_1:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[IDXPROM1:%.*]] = zext i32 [[WORKITEM_ID_X_1]] to i64 +; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[P]] to ptr addrspace(5) +; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds float, ptr addrspace(5) [[TMP1]], i64 [[IDXPROM1]] +; CHECK-NEXT: [[LOAD1:%.*]] = load float, ptr addrspace(5) [[ARRAYIDX1]], align 4 +; CHECK-NEXT: [[ADD1:%.*]] = fadd float [[LOAD1]], 4.000000e+00 +; CHECK-NEXT: br label %[[IF_END]] +; CHECK: [[IF_END]]: +; CHECK-NEXT: [[PHI:%.*]] = phi float [ [[ADD0]], %[[IF_THEN_SHARED]] ], [ [[ADD1]], %[[IF_THEN_PRIVATE]] ] +; CHECK-NEXT: ret float [[PHI]] +; +entry: + %tobool.not = icmp eq i32 %c, 0 + br i1 %tobool.not, label %if.then.shared, label %if.then.private + +if.then.shared: + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %p) + tail call void @llvm.assume(i1 %is.shared) + %workitem.id.x.0 = tail call i32 @llvm.amdgcn.workitem.id.x() + %idxprom = zext i32 %workitem.id.x.0 to i64 + %arrayidx0 = getelementptr inbounds float, ptr %p, i64 %idxprom + %load0 = load float, ptr %arrayidx0, align 4 + %add0 = fadd float %load0, 4.0 + br label %if.end + +if.then.private: + %is.private = call i1 @llvm.amdgcn.is.private(ptr %p) + tail call void @llvm.assume(i1 %is.private) + %workitem.id.x.1 = tail call i32 @llvm.amdgcn.workitem.id.x() + %idxprom1 = zext i32 %workitem.id.x.1 to i64 + %arrayidx1 = getelementptr inbounds float, ptr %p, i64 %idxprom1 + %load1 = load float, ptr %arrayidx1, align 4 + %add1 = fadd float %load1, 4.0 + br label %if.end + +if.end: + %phi = phi float [ %add0, %if.then.shared ], [ %add1, %if.then.private ] + ret float %phi +} + +define float @assume_func_arg_is_shared_load(ptr %flat.ptr) { +; CHECK-LABEL: define float @assume_func_arg_is_shared_load( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) { +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[FLAT_PTR]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_SHARED]]) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr [[FLAT_PTR]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %flat.ptr) + tail call void @llvm.assume(i1 %is.shared) + %load = load float, ptr %flat.ptr, align 4 + ret float %load +} + +define float @assume_func_arg_is_private_load(ptr %flat.ptr) { +; CHECK-LABEL: define float @assume_func_arg_is_private_load( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) { +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[FLAT_PTR]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_PRIVATE]]) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr [[FLAT_PTR]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.private = call i1 @llvm.amdgcn.is.private(ptr %flat.ptr) + tail call void @llvm.assume(i1 %is.private) + %load = load float, ptr %flat.ptr, align 4 + ret float %load +} + +define float @assume_func_arg_is_not_shared_not_private(ptr %flat.ptr) { +; CHECK-LABEL: define float @assume_func_arg_is_not_shared_not_private( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) { +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[FLAT_PTR]]) +; CHECK-NEXT: [[NOT_PRIVATE:%.*]] = xor i1 [[IS_PRIVATE]], true +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[FLAT_PTR]]) +; CHECK-NEXT: [[NOT_SHARED:%.*]] = xor i1 [[IS_SHARED]], true +; CHECK-NEXT: [[NOT_PRIVATE_AND_NOT_SHARED:%.*]] = and i1 [[NOT_PRIVATE]], [[NOT_SHARED]] +; CHECK-NEXT: tail call void @llvm.assume(i1 [[NOT_PRIVATE_AND_NOT_SHARED]]) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr [[FLAT_PTR]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.private = call i1 @llvm.amdgcn.is.private(ptr %flat.ptr) + %not.private = xor i1 %is.private, true + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %flat.ptr) + %not.shared = xor i1 %is.shared, true + %not.private.and.not.shared = and i1 %not.private, %not.shared + tail call void @llvm.assume(i1 %not.private.and.not.shared) + %load = load float, ptr %flat.ptr, align 4 + ret float %load +} + +define float @assume_func_arg_is_not_private_load(ptr %flat.ptr) { +; CHECK-LABEL: define float @assume_func_arg_is_not_private_load( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]]) { +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[FLAT_PTR]]) +; CHECK-NEXT: [[NOT_IS_PRIVATE:%.*]] = xor i1 [[IS_PRIVATE]], true +; CHECK-NEXT: tail call void @llvm.assume(i1 [[NOT_IS_PRIVATE]]) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr [[FLAT_PTR]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.private = call i1 @llvm.amdgcn.is.private(ptr %flat.ptr) + %not.is.private = xor i1 %is.private, true + tail call void @llvm.assume(i1 %not.is.private) + %load = load float, ptr %flat.ptr, align 4 + ret float %load +} + +define i64 @assume_func_arg_is_not_private_atomicrmw(ptr %flat.ptr, i64 %val) { +; CHECK-LABEL: define i64 @assume_func_arg_is_not_private_atomicrmw( +; CHECK-SAME: ptr [[FLAT_PTR:%.*]], i64 [[VAL:%.*]]) { +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[FLAT_PTR]]) +; CHECK-NEXT: [[NOT_IS_PRIVATE:%.*]] = xor i1 [[IS_PRIVATE]], true +; CHECK-NEXT: tail call void @llvm.assume(i1 [[NOT_IS_PRIVATE]]) +; CHECK-NEXT: [[RMW:%.*]] = atomicrmw sub ptr [[FLAT_PTR]], i64 [[VAL]] seq_cst, align 4 +; CHECK-NEXT: ret i64 [[RMW]] +; + %is.private = call i1 @llvm.amdgcn.is.private(ptr %flat.ptr) + %not.is.private = xor i1 %is.private, true + tail call void @llvm.assume(i1 %not.is.private) + %rmw = atomicrmw sub ptr %flat.ptr, i64 %val seq_cst, align 4 + ret i64 %rmw +} + +define float @contradictory_assume_after_gep_same_block(ptr %p) { +; CHECK-LABEL: define float @contradictory_assume_after_gep_same_block( +; CHECK-SAME: ptr [[P:%.*]]) { +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[P]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_SHARED]]) +; CHECK-NEXT: [[WORKITEM_ID_X:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[WORKITEM_ID_X]] to i64 +; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[P]] to ptr addrspace(3) +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr addrspace(3) [[TMP1]], i64 [[IDXPROM]] +; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[P]], i64 [[IDXPROM]] +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[TMP2]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 false) +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr addrspace(3) [[GEP]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %p) + tail call void @llvm.assume(i1 %is.shared) + %workitem.id.x = tail call i32 @llvm.amdgcn.workitem.id.x() + %idxprom = zext i32 %workitem.id.x to i64 + %gep = getelementptr inbounds float, ptr %p, i64 %idxprom + %is.private = call i1 @llvm.amdgcn.is.private(ptr %gep) + tail call void @llvm.assume(i1 %is.private) + %load = load float, ptr %gep, align 4 + ret float %load +} + +define float @contradictory_assume_argument_same_block(ptr %p) { +; CHECK-LABEL: define float @contradictory_assume_argument_same_block( +; CHECK-SAME: ptr [[P:%.*]]) { +; CHECK-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[P]]) +; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[P]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_SHARED]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[IS_PRIVATE]]) +; CHECK-NEXT: [[WORKITEM_ID_X:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[WORKITEM_ID_X]] to i64 +; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[P]] to ptr addrspace(3) +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr addrspace(3) [[TMP1]], i64 [[IDXPROM]] +; CHECK-NEXT: [[LOAD:%.*]] = load float, ptr addrspace(3) [[GEP]], align 4 +; CHECK-NEXT: ret float [[LOAD]] +; + %is.shared = call i1 @llvm.amdgcn.is.shared(ptr %p) + %is.private = call i1 @llvm.amdgcn.is.private(ptr %p) + tail call void @llvm.assume(i1 %is.shared) + tail call void @llvm.assume(i1 %is.private) + %workitem.id.x = tail call i32 @llvm.amdgcn.workitem.id.x() + %idxprom = zext i32 %workitem.id.x to i64 + %gep = getelementptr inbounds float, ptr %p, i64 %idxprom + %load = load float, ptr %gep, align 4 + ret float %load +} + declare void @llvm.assume(i1) declare i1 @llvm.amdgcn.is.shared(ptr nocapture) declare i1 @llvm.amdgcn.is.private(ptr nocapture) 

[darkbuck]

LGTM