[X86] Manage atomic load of fp -> int promotion in DAG

[jofrn]

When lowering atomic <1 x T> vector types with floats, selection can fail since
this pattern is unsupported. To support this, floats can be casted to
an integer type of the same size.

[jofrn]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• [AtomicExpand] Add bitcasts when expanding load atomic vector #148900
• [X86] Cast atomic vectors in IR to support floats #148899
• [X86] Remove extra MOV after widening atomic load #148898
• [SelectionDAG] Widen <2 x T> vector types for atomic load #148897
• [X86] Add atomic vector tests for unaligned >1 sizes. #148896
• [X86] Manage atomic load of fp -> int promotion in DAG #148895 👈 (View in Graphite)
• [SelectionDAG] Legalize <1 x T> vector types for atomic load #148894
• IR/Verifier: Allow vector type in atomic load and store #148893
• main

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

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: None (jofrn)

Changes

When lowering atomic <1 x T> vector types with floats, selection can fail since
this pattern is unsupported. To support this, floats can be casted to
an integer type of the same size.

---

Full diff: https://github.com/llvm/llvm-project/pull/148895.diff

2 Files Affected:

• (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+4)
• (modified) llvm/test/CodeGen/X86/atomic-load-store.ll (+117)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index b1a3e3c006bb3..776d3c0a42e2f 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -2653,6 +2653,10 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
         setOperationAction(Op, MVT::f32, Promote);
   }
 
+  setOperationPromotedToType(ISD::ATOMIC_LOAD, MVT::f16, MVT::i16);
+  setOperationPromotedToType(ISD::ATOMIC_LOAD, MVT::f32, MVT::i32);
+  setOperationPromotedToType(ISD::ATOMIC_LOAD, MVT::f64, MVT::i64);
+
   // We have target-specific dag combine patterns for the following nodes:
   setTargetDAGCombine({ISD::VECTOR_SHUFFLE,
                        ISD::SCALAR_TO_VECTOR,
diff --git a/llvm/test/CodeGen/X86/atomic-load-store.ll b/llvm/test/CodeGen/X86/atomic-load-store.ll
index 4f5cb5a4e9247..9fab8b98b4af0 100644
--- a/llvm/test/CodeGen/X86/atomic-load-store.ll
+++ b/llvm/test/CodeGen/X86/atomic-load-store.ll
@@ -269,3 +269,120 @@ define <1 x i64> @atomic_vec1_i64_align(ptr %x) nounwind {
   %ret = load atomic <1 x i64>, ptr %x acquire, align 8
   ret <1 x i64> %ret
 }
+
+define <1 x half> @atomic_vec1_half(ptr %x) {
+; CHECK-O3-LABEL: atomic_vec1_half:
+; CHECK-O3:       # %bb.0:
+; CHECK-O3-NEXT:    movzwl (%rdi), %eax
+; CHECK-O3-NEXT:    pinsrw $0, %eax, %xmm0
+; CHECK-O3-NEXT:    retq
+;
+; CHECK-SSE-O3-LABEL: atomic_vec1_half:
+; CHECK-SSE-O3:       # %bb.0:
+; CHECK-SSE-O3-NEXT:    movzwl (%rdi), %eax
+; CHECK-SSE-O3-NEXT:    pinsrw $0, %eax, %xmm0
+; CHECK-SSE-O3-NEXT:    retq
+;
+; CHECK-AVX-O3-LABEL: atomic_vec1_half:
+; CHECK-AVX-O3:       # %bb.0:
+; CHECK-AVX-O3-NEXT:    movzwl (%rdi), %eax
+; CHECK-AVX-O3-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
+; CHECK-AVX-O3-NEXT:    retq
+;
+; CHECK-O0-LABEL: atomic_vec1_half:
+; CHECK-O0:       # %bb.0:
+; CHECK-O0-NEXT:    movw (%rdi), %cx
+; CHECK-O0-NEXT:    # implicit-def: $eax
+; CHECK-O0-NEXT:    movw %cx, %ax
+; CHECK-O0-NEXT:    # implicit-def: $xmm0
+; CHECK-O0-NEXT:    pinsrw $0, %eax, %xmm0
+; CHECK-O0-NEXT:    retq
+;
+; CHECK-SSE-O0-LABEL: atomic_vec1_half:
+; CHECK-SSE-O0:       # %bb.0:
+; CHECK-SSE-O0-NEXT:    movw (%rdi), %cx
+; CHECK-SSE-O0-NEXT:    # implicit-def: $eax
+; CHECK-SSE-O0-NEXT:    movw %cx, %ax
+; CHECK-SSE-O0-NEXT:    # implicit-def: $xmm0
+; CHECK-SSE-O0-NEXT:    pinsrw $0, %eax, %xmm0
+; CHECK-SSE-O0-NEXT:    retq
+;
+; CHECK-AVX-O0-LABEL: atomic_vec1_half:
+; CHECK-AVX-O0:       # %bb.0:
+; CHECK-AVX-O0-NEXT:    movw (%rdi), %cx
+; CHECK-AVX-O0-NEXT:    # implicit-def: $eax
+; CHECK-AVX-O0-NEXT:    movw %cx, %ax
+; CHECK-AVX-O0-NEXT:    # implicit-def: $xmm0
+; CHECK-AVX-O0-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
+; CHECK-AVX-O0-NEXT:    retq
+  %ret = load atomic <1 x half>, ptr %x acquire, align 2
+  ret <1 x half> %ret
+}
+
+define <1 x float> @atomic_vec1_float(ptr %x) {
+; CHECK-O3-LABEL: atomic_vec1_float:
+; CHECK-O3:       # %bb.0:
+; CHECK-O3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; CHECK-O3-NEXT:    retq
+;
+; CHECK-SSE-O3-LABEL: atomic_vec1_float:
+; CHECK-SSE-O3:       # %bb.0:
+; CHECK-SSE-O3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; CHECK-SSE-O3-NEXT:    retq
+;
+; CHECK-AVX-O3-LABEL: atomic_vec1_float:
+; CHECK-AVX-O3:       # %bb.0:
+; CHECK-AVX-O3-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; CHECK-AVX-O3-NEXT:    retq
+;
+; CHECK-O0-LABEL: atomic_vec1_float:
+; CHECK-O0:       # %bb.0:
+; CHECK-O0-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; CHECK-O0-NEXT:    retq
+;
+; CHECK-SSE-O0-LABEL: atomic_vec1_float:
+; CHECK-SSE-O0:       # %bb.0:
+; CHECK-SSE-O0-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; CHECK-SSE-O0-NEXT:    retq
+;
+; CHECK-AVX-O0-LABEL: atomic_vec1_float:
+; CHECK-AVX-O0:       # %bb.0:
+; CHECK-AVX-O0-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; CHECK-AVX-O0-NEXT:    retq
+  %ret = load atomic <1 x float>, ptr %x acquire, align 4
+  ret <1 x float> %ret
+}
+
+define <1 x double> @atomic_vec1_double_align(ptr %x) nounwind {
+; CHECK-O3-LABEL: atomic_vec1_double_align:
+; CHECK-O3:       # %bb.0:
+; CHECK-O3-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK-O3-NEXT:    retq
+;
+; CHECK-SSE-O3-LABEL: atomic_vec1_double_align:
+; CHECK-SSE-O3:       # %bb.0:
+; CHECK-SSE-O3-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK-SSE-O3-NEXT:    retq
+;
+; CHECK-AVX-O3-LABEL: atomic_vec1_double_align:
+; CHECK-AVX-O3:       # %bb.0:
+; CHECK-AVX-O3-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK-AVX-O3-NEXT:    retq
+;
+; CHECK-O0-LABEL: atomic_vec1_double_align:
+; CHECK-O0:       # %bb.0:
+; CHECK-O0-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK-O0-NEXT:    retq
+;
+; CHECK-SSE-O0-LABEL: atomic_vec1_double_align:
+; CHECK-SSE-O0:       # %bb.0:
+; CHECK-SSE-O0-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK-SSE-O0-NEXT:    retq
+;
+; CHECK-AVX-O0-LABEL: atomic_vec1_double_align:
+; CHECK-AVX-O0:       # %bb.0:
+; CHECK-AVX-O0-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK-AVX-O0-NEXT:    retq
+  %ret = load atomic <1 x double>, ptr %x acquire, align 8
+  ret <1 x double> %ret
+}