[llvm] [RISCV] Fix incorrect stride computation in concat_vectors-of-loads combine (PR #217401)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Aug 19 10:37:57 PDT 2026
llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-risc-v
Author: Craig Topper (topperc)
<details>
<summary>Changes</summary>
performCONCAT_VECTORSCombine's computed the stride between two loads as BIO2.getOffset() - BIO1.getOffset(), but getOffset() only returns each load's own trailing offset component, not the base-to-base displacement. For loads based on two distinct fixed FrameIndex objects (e.g. an illegal, non-power-of-2 vector argument scalarized into per-element stack slots), this always computed 0 regardless of the true distance between the slots, since equalBaseIndex computes the correct relative offset separately and returns it via an out-parameter that was being discarded.
Use the 3 argument equalBaseIndex overload and its computed offset instead.
Fixes #<!-- -->217369.
Assisted-by: Claude
---
Full diff: https://github.com/llvm/llvm-project/pull/217401.diff
2 Files Affected:
- (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+3-2)
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-strided-load-combine.ll (+350)
``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 44359a9e6155f..7f89c348ed4a5 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -21976,8 +21976,9 @@ static SDValue performCONCAT_VECTORSCombine(SDNode *N, SelectionDAG &DAG,
// common constant stride, then return the constant stride.
BaseIndexOffset BIO1 = BaseIndexOffset::match(Ld1, DAG);
BaseIndexOffset BIO2 = BaseIndexOffset::match(Ld2, DAG);
- if (BIO1.equalBaseIndex(BIO2, DAG))
- return {{BIO2.getOffset() - BIO1.getOffset(), false}};
+ int64_t PtrDiff;
+ if (BIO1.equalBaseIndex(BIO2, DAG, PtrDiff))
+ return {{PtrDiff, false}};
// Otherwise try to match (add LastPtr, Stride) or (add NextPtr, Stride)
SDValue P1 = Ld1->getBasePtr();
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-strided-load-combine.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-strided-load-combine.ll
index a6cc590deab64..2cadb0b94ad58 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-strided-load-combine.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-strided-load-combine.ll
@@ -594,3 +594,353 @@ define <16 x i8> @widen_4xv4i8_immediate_expand(ptr %p, i64 %s) {
%abcd = shufflevector <16 x i8> %abcx, <16 x i8> %dx, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
ret <16 x i8> %abcd
}
+
+; An illegal, non-power-of-2 vector argument gets scalarized into a sequence
+; of loads from distinct fixed (incoming-argument) stack slots, which are
+; then reassembled with concat_vectors. The base pointers of these loads are
+; different FrameIndex nodes, so computing their pointer difference relies on
+; BaseIndexOffset::equalBaseIndex's fixed-object handling rather than the
+; loads sharing a common base. This used to compute a stride of 0 for every
+; pair (instead of the correct 4), which produced a load whose MachineMemOperand
+; was undersized for its memory VT and tripped an assertion in MemSDNode.
+define <33 x i32> @concat_loads_of_fixed_stack_args(<33 x i32> %a, <33 x i32> %b) {
+; RV32-LABEL: concat_loads_of_fixed_stack_args:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vslideup.vi v18, v19, 1
+; RV32-NEXT: vslideup.vi v16, v17, 1
+; RV32-NEXT: vslideup.vi v22, v23, 1
+; RV32-NEXT: vslideup.vi v20, v21, 1
+; RV32-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT: vslideup.vi v16, v18, 2
+; RV32-NEXT: vslideup.vi v20, v22, 2
+; RV32-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV32-NEXT: vslideup.vi v16, v20, 4
+; RV32-NEXT: addi a2, a1, 60
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v22, (a2)
+; RV32-NEXT: addi a2, a1, 56
+; RV32-NEXT: vle32.v v21, (a2)
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vslideup.vi v10, v11, 1
+; RV32-NEXT: vslideup.vi v8, v9, 1
+; RV32-NEXT: addi a2, a1, 52
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v23, (a2)
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vslideup.vi v14, v15, 1
+; RV32-NEXT: vslideup.vi v12, v13, 1
+; RV32-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT: vslideup.vi v8, v10, 2
+; RV32-NEXT: addi a2, a1, 48
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v20, (a2)
+; RV32-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT: vslideup.vi v12, v14, 2
+; RV32-NEXT: addi a2, a1, 44
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v14, (a2)
+; RV32-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV32-NEXT: vslideup.vi v8, v12, 4
+; RV32-NEXT: addi a2, a1, 40
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v13, (a2)
+; RV32-NEXT: addi a2, a1, 36
+; RV32-NEXT: vle32.v v15, (a2)
+; RV32-NEXT: vsetivli zero, 16, e32, m4, ta, ma
+; RV32-NEXT: vslideup.vi v8, v16, 8
+; RV32-NEXT: addi a2, a1, 32
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v12, (a2)
+; RV32-NEXT: addi a2, a1, 28
+; RV32-NEXT: vle32.v v17, (a2)
+; RV32-NEXT: addi a2, a1, 24
+; RV32-NEXT: vle32.v v18, (a2)
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vslideup.vi v13, v14, 1
+; RV32-NEXT: vslideup.vi v12, v15, 1
+; RV32-NEXT: addi a2, a1, 20
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v15, (a2)
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vslideup.vi v21, v22, 1
+; RV32-NEXT: vslideup.vi v20, v23, 1
+; RV32-NEXT: addi a2, a1, 12
+; RV32-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT: vslideup.vi v12, v13, 2
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v14, (a2)
+; RV32-NEXT: addi a2, a1, 8
+; RV32-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT: vslideup.vi v20, v21, 2
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v19, (a2)
+; RV32-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV32-NEXT: vslideup.vi v12, v20, 4
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v16, (a1)
+; RV32-NEXT: addi a2, a1, 4
+; RV32-NEXT: vle32.v v20, (a2)
+; RV32-NEXT: addi a2, a1, 16
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vslideup.vi v19, v14, 1
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v14, (a2)
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vslideup.vi v16, v20, 1
+; RV32-NEXT: vslideup.vi v18, v17, 1
+; RV32-NEXT: vslideup.vi v14, v15, 1
+; RV32-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT: vslideup.vi v16, v19, 2
+; RV32-NEXT: vslideup.vi v14, v18, 2
+; RV32-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV32-NEXT: vslideup.vi v16, v14, 4
+; RV32-NEXT: vsetivli zero, 16, e32, m4, ta, ma
+; RV32-NEXT: vslideup.vi v16, v12, 8
+; RV32-NEXT: li a2, 32
+; RV32-NEXT: addi a3, sp, 40
+; RV32-NEXT: vsetvli zero, a2, e32, m8, ta, ma
+; RV32-NEXT: vslideup.vi v8, v16, 16
+; RV32-NEXT: vle32.v v16, (a3)
+; RV32-NEXT: vadd.vv v8, v8, v16
+; RV32-NEXT: addi a1, a1, 64
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v16, (a1)
+; RV32-NEXT: vsetvli zero, a2, e32, m8, ta, ma
+; RV32-NEXT: vse32.v v8, (a0)
+; RV32-NEXT: addi a1, sp, 168
+; RV32-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV32-NEXT: vle32.v v8, (a1)
+; RV32-NEXT: vadd.vv v8, v16, v8
+; RV32-NEXT: addi a0, a0, 128
+; RV32-NEXT: vse32.v v8, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: concat_loads_of_fixed_stack_args:
+; RV64: # %bb.0:
+; RV64-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV64-NEXT: vslideup.vi v18, v19, 1
+; RV64-NEXT: vslideup.vi v16, v17, 1
+; RV64-NEXT: vslideup.vi v22, v23, 1
+; RV64-NEXT: vslideup.vi v20, v21, 1
+; RV64-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV64-NEXT: vslideup.vi v16, v18, 2
+; RV64-NEXT: vslideup.vi v20, v22, 2
+; RV64-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64-NEXT: vslideup.vi v16, v20, 4
+; RV64-NEXT: addi a2, a1, 60
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v22, (a2)
+; RV64-NEXT: addi a2, a1, 56
+; RV64-NEXT: vle32.v v21, (a2)
+; RV64-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV64-NEXT: vslideup.vi v10, v11, 1
+; RV64-NEXT: vslideup.vi v8, v9, 1
+; RV64-NEXT: addi a2, a1, 52
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v23, (a2)
+; RV64-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV64-NEXT: vslideup.vi v14, v15, 1
+; RV64-NEXT: vslideup.vi v12, v13, 1
+; RV64-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV64-NEXT: vslideup.vi v8, v10, 2
+; RV64-NEXT: addi a2, a1, 48
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v20, (a2)
+; RV64-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV64-NEXT: vslideup.vi v12, v14, 2
+; RV64-NEXT: addi a2, a1, 44
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v14, (a2)
+; RV64-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64-NEXT: vslideup.vi v8, v12, 4
+; RV64-NEXT: addi a2, a1, 40
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v13, (a2)
+; RV64-NEXT: addi a2, a1, 36
+; RV64-NEXT: vle32.v v15, (a2)
+; RV64-NEXT: vsetivli zero, 16, e32, m4, ta, ma
+; RV64-NEXT: vslideup.vi v8, v16, 8
+; RV64-NEXT: addi a2, a1, 32
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v12, (a2)
+; RV64-NEXT: addi a2, a1, 28
+; RV64-NEXT: vle32.v v17, (a2)
+; RV64-NEXT: addi a2, a1, 24
+; RV64-NEXT: vle32.v v18, (a2)
+; RV64-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV64-NEXT: vslideup.vi v13, v14, 1
+; RV64-NEXT: vslideup.vi v12, v15, 1
+; RV64-NEXT: addi a2, a1, 20
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v15, (a2)
+; RV64-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV64-NEXT: vslideup.vi v21, v22, 1
+; RV64-NEXT: vslideup.vi v20, v23, 1
+; RV64-NEXT: addi a2, a1, 12
+; RV64-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV64-NEXT: vslideup.vi v12, v13, 2
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v14, (a2)
+; RV64-NEXT: addi a2, a1, 8
+; RV64-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV64-NEXT: vslideup.vi v20, v21, 2
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v19, (a2)
+; RV64-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64-NEXT: vslideup.vi v12, v20, 4
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v16, (a1)
+; RV64-NEXT: addi a2, a1, 4
+; RV64-NEXT: vle32.v v20, (a2)
+; RV64-NEXT: addi a2, a1, 16
+; RV64-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV64-NEXT: vslideup.vi v19, v14, 1
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v14, (a2)
+; RV64-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV64-NEXT: vslideup.vi v16, v20, 1
+; RV64-NEXT: vslideup.vi v18, v17, 1
+; RV64-NEXT: vslideup.vi v14, v15, 1
+; RV64-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV64-NEXT: vslideup.vi v16, v19, 2
+; RV64-NEXT: vslideup.vi v14, v18, 2
+; RV64-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64-NEXT: vslideup.vi v16, v14, 4
+; RV64-NEXT: vsetivli zero, 16, e32, m4, ta, ma
+; RV64-NEXT: vslideup.vi v16, v12, 8
+; RV64-NEXT: li a2, 32
+; RV64-NEXT: addi a3, sp, 40
+; RV64-NEXT: vsetvli zero, a2, e32, m8, ta, ma
+; RV64-NEXT: vslideup.vi v8, v16, 16
+; RV64-NEXT: vle32.v v16, (a3)
+; RV64-NEXT: vadd.vv v8, v8, v16
+; RV64-NEXT: addi a1, a1, 64
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v16, (a1)
+; RV64-NEXT: vsetvli zero, a2, e32, m8, ta, ma
+; RV64-NEXT: vse32.v v8, (a0)
+; RV64-NEXT: addi a1, sp, 168
+; RV64-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; RV64-NEXT: vle32.v v8, (a1)
+; RV64-NEXT: vadd.vv v8, v16, v8
+; RV64-NEXT: addi a0, a0, 128
+; RV64-NEXT: vsetivli zero, 1, e32, m1, ta, ma
+; RV64-NEXT: vse32.v v8, (a0)
+; RV64-NEXT: ret
+;
+; ZVE64F-LABEL: concat_loads_of_fixed_stack_args:
+; ZVE64F: # %bb.0:
+; ZVE64F-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v18, v19, 1
+; ZVE64F-NEXT: vslideup.vi v16, v17, 1
+; ZVE64F-NEXT: vslideup.vi v22, v23, 1
+; ZVE64F-NEXT: vslideup.vi v20, v21, 1
+; ZVE64F-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; ZVE64F-NEXT: vslideup.vi v16, v18, 2
+; ZVE64F-NEXT: vslideup.vi v20, v22, 2
+; ZVE64F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v16, v20, 4
+; ZVE64F-NEXT: addi a2, a1, 60
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v22, (a2)
+; ZVE64F-NEXT: addi a2, a1, 56
+; ZVE64F-NEXT: vle32.v v21, (a2)
+; ZVE64F-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v10, v11, 1
+; ZVE64F-NEXT: vslideup.vi v8, v9, 1
+; ZVE64F-NEXT: addi a2, a1, 52
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v23, (a2)
+; ZVE64F-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v14, v15, 1
+; ZVE64F-NEXT: vslideup.vi v12, v13, 1
+; ZVE64F-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; ZVE64F-NEXT: vslideup.vi v8, v10, 2
+; ZVE64F-NEXT: addi a2, a1, 48
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v20, (a2)
+; ZVE64F-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; ZVE64F-NEXT: vslideup.vi v12, v14, 2
+; ZVE64F-NEXT: addi a2, a1, 44
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v14, (a2)
+; ZVE64F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v8, v12, 4
+; ZVE64F-NEXT: addi a2, a1, 40
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v13, (a2)
+; ZVE64F-NEXT: addi a2, a1, 36
+; ZVE64F-NEXT: vle32.v v15, (a2)
+; ZVE64F-NEXT: vsetivli zero, 16, e32, m4, ta, ma
+; ZVE64F-NEXT: vslideup.vi v8, v16, 8
+; ZVE64F-NEXT: addi a2, a1, 32
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v12, (a2)
+; ZVE64F-NEXT: addi a2, a1, 28
+; ZVE64F-NEXT: vle32.v v17, (a2)
+; ZVE64F-NEXT: addi a2, a1, 24
+; ZVE64F-NEXT: vle32.v v18, (a2)
+; ZVE64F-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v13, v14, 1
+; ZVE64F-NEXT: vslideup.vi v12, v15, 1
+; ZVE64F-NEXT: addi a2, a1, 20
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v15, (a2)
+; ZVE64F-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v21, v22, 1
+; ZVE64F-NEXT: vslideup.vi v20, v23, 1
+; ZVE64F-NEXT: addi a2, a1, 12
+; ZVE64F-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; ZVE64F-NEXT: vslideup.vi v12, v13, 2
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v14, (a2)
+; ZVE64F-NEXT: addi a2, a1, 8
+; ZVE64F-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; ZVE64F-NEXT: vslideup.vi v20, v21, 2
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v19, (a2)
+; ZVE64F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v12, v20, 4
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v16, (a1)
+; ZVE64F-NEXT: addi a2, a1, 4
+; ZVE64F-NEXT: vle32.v v20, (a2)
+; ZVE64F-NEXT: addi a2, a1, 16
+; ZVE64F-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v19, v14, 1
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v14, (a2)
+; ZVE64F-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v16, v20, 1
+; ZVE64F-NEXT: vslideup.vi v18, v17, 1
+; ZVE64F-NEXT: vslideup.vi v14, v15, 1
+; ZVE64F-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; ZVE64F-NEXT: vslideup.vi v16, v19, 2
+; ZVE64F-NEXT: vslideup.vi v14, v18, 2
+; ZVE64F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; ZVE64F-NEXT: vslideup.vi v16, v14, 4
+; ZVE64F-NEXT: vsetivli zero, 16, e32, m4, ta, ma
+; ZVE64F-NEXT: vslideup.vi v16, v12, 8
+; ZVE64F-NEXT: li a2, 32
+; ZVE64F-NEXT: addi a3, sp, 40
+; ZVE64F-NEXT: vsetvli zero, a2, e32, m8, ta, ma
+; ZVE64F-NEXT: vslideup.vi v8, v16, 16
+; ZVE64F-NEXT: vle32.v v16, (a3)
+; ZVE64F-NEXT: vadd.vv v8, v8, v16
+; ZVE64F-NEXT: addi a1, a1, 64
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v16, (a1)
+; ZVE64F-NEXT: vsetvli zero, a2, e32, m8, ta, ma
+; ZVE64F-NEXT: vse32.v v8, (a0)
+; ZVE64F-NEXT: addi a1, sp, 168
+; ZVE64F-NEXT: vsetivli zero, 1, e32, mf2, ta, ma
+; ZVE64F-NEXT: vle32.v v8, (a1)
+; ZVE64F-NEXT: vadd.vv v8, v16, v8
+; ZVE64F-NEXT: addi a0, a0, 128
+; ZVE64F-NEXT: vsetivli zero, 1, e32, m1, ta, ma
+; ZVE64F-NEXT: vse32.v v8, (a0)
+; ZVE64F-NEXT: ret
+ %r = add <33 x i32> %a, %b
+ ret <33 x i32> %r
+}
``````````
</details>
https://github.com/llvm/llvm-project/pull/217401
More information about the llvm-commits
mailing list