[llvm] [SDAG] Read-only intrinsics must have WillReturn attribute to be treated as loads (PR #99999)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-selectiondag

Author: Kevin McAfee (kalxr)

<details>
<summary>Changes</summary>

This change avoids deleting `!willReturn` intrinsics for which the return value is unused when building the SDAG. Currently, calls to read-only intrinsics not marked with `IntrWillReturn` cannot deleted at the LLVM IR level but may be deleted when building the SDAG. This appears incorrect. These calls are unsafe to remove from the IR because the functions are `!willReturn` and should also be unsafe to remove from the SDAG for the same reason. This change aligns the behavior of the SDAG to that of LLVM IR.

Additionally, update tests where instruction order changed slightly as a result of this change.

---
Full diff: https://github.com/llvm/llvm-project/pull/99999.diff


3 Files Affected:

- (modified) llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp (+1-1) 
- (modified) llvm/test/CodeGen/BPF/intrinsics.ll (+4-4) 
- (modified) llvm/test/CodeGen/BPF/objdump_intrinsics.ll (+4-4) 


``````````diff
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 37b1131d2f8a3..29e6466b12cde 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -5227,7 +5227,7 @@ void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I,
   // definition.
   const Function *F = I.getCalledFunction();
   bool HasChain = !F->doesNotAccessMemory();
-  bool OnlyLoad = HasChain && F->onlyReadsMemory();
+  bool OnlyLoad = HasChain && F->onlyReadsMemory() && F->willReturn();
 
   // Build the operand list.
   SmallVector<SDValue, 8> Ops;
diff --git a/llvm/test/CodeGen/BPF/intrinsics.ll b/llvm/test/CodeGen/BPF/intrinsics.ll
index 0f59fd5604732..8331ee5ff04ee 100644
--- a/llvm/test/CodeGen/BPF/intrinsics.ll
+++ b/llvm/test/CodeGen/BPF/intrinsics.ll
@@ -31,10 +31,10 @@ define i32 @ld_h(ptr %ctx, ptr %ctx2, i32 %foo) #0 {
   %5 = trunc i64 %4 to i32
   ret i32 %5
 ; CHECK-LABEL: ld_h:
-; CHECK-EL: r0 = *(u16 *)skb[r
 ; CHECK-EL: r0 = *(u16 *)skb[123]
-; CHECK-EB: r0 = *(u16 *)skb[r
+; CHECK-EL: r0 = *(u16 *)skb[r
 ; CHECK-EB: r0 = *(u16 *)skb[123]
+; CHECK-EB: r0 = *(u16 *)skb[r
 }
 
 declare i64 @llvm.bpf.load.half(ptr, i64) #1
@@ -48,10 +48,10 @@ define i32 @ld_w(ptr %ctx, ptr %ctx2, i32 %foo) #0 {
   %5 = trunc i64 %4 to i32
   ret i32 %5
 ; CHECK-LABEL: ld_w:
-; CHECK-EL: r0 = *(u32 *)skb[r
 ; CHECK-EL: r0 = *(u32 *)skb[123]
-; CHECK-EB: r0 = *(u32 *)skb[r
+; CHECK-EL: r0 = *(u32 *)skb[r
 ; CHECK-EB: r0 = *(u32 *)skb[123]
+; CHECK-EB: r0 = *(u32 *)skb[r
 }
 
 declare i64 @llvm.bpf.load.word(ptr, i64) #1
diff --git a/llvm/test/CodeGen/BPF/objdump_intrinsics.ll b/llvm/test/CodeGen/BPF/objdump_intrinsics.ll
index 92db0882398b6..5497fce989d21 100644
--- a/llvm/test/CodeGen/BPF/objdump_intrinsics.ll
+++ b/llvm/test/CodeGen/BPF/objdump_intrinsics.ll
@@ -31,10 +31,10 @@ define i32 @ld_h(ptr %ctx, ptr %ctx2, i32 %foo) #0 {
   %5 = trunc i64 %4 to i32
   ret i32 %5
 ; CHECK-LABEL: ld_h:
-; CHECK-EL: r0 = *(u16 *)skb[r
 ; CHECK-EL: r0 = *(u16 *)skb[123]
-; CHECK-EB: r0 = *(u16 *)skb[r
+; CHECK-EL: r0 = *(u16 *)skb[r
 ; CHECK-EB: r0 = *(u16 *)skb[123]
+; CHECK-EB: r0 = *(u16 *)skb[r
 }
 
 declare i64 @llvm.bpf.load.half(ptr, i64) #1
@@ -48,10 +48,10 @@ define i32 @ld_w(ptr %ctx, ptr %ctx2, i32 %foo) #0 {
   %5 = trunc i64 %4 to i32
   ret i32 %5
 ; CHECK-LABEL: ld_w:
-; CHECK-EL: r0 = *(u32 *)skb[r
 ; CHECK-EL: r0 = *(u32 *)skb[123]
-; CHECK-EB: r0 = *(u32 *)skb[r
+; CHECK-EL: r0 = *(u32 *)skb[r
 ; CHECK-EB: r0 = *(u32 *)skb[123]
+; CHECK-EB: r0 = *(u32 *)skb[r
 }
 
 declare i64 @llvm.bpf.load.word(ptr, i64) #1

``````````

</details>


https://github.com/llvm/llvm-project/pull/99999