Skip to content

[RISCV] Fix misuse of EVT::changeVectorElementType() in legalizeScatterGatherIndexType. #165829

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Oct 31, 2025
Merged

[RISCV] Fix misuse of EVT::changeVectorElementType() in legalizeScatterGatherIndexType. #165829

merged 1 commit into from
Oct 31, 2025

Conversation

@topperc
Copy link
Collaborator

@topperc topperc commented Oct 31, 2025

This function doesn't work well when the type is a SimpleVT, but the changed type isn't. We need an LLVMContext to make an non-SimpleVT, but there's nowhere to get it from.

Fix this by using EVT::getVectorVT instead.

In the added test, v7i8 is a SimpleVT, but v7i64 isn't.

@topperc
[RISCV] Fix misuse of EVT::changeVectorElementType() in legalizeScatt…
bae211a 
…erGatherIndexType.

This function doesn't work well when the type is a SimpleVT, but
the changed type isn't. We need an LLVMContext to make an non-SimpleVT,
but there's nowhere to get it from.

Fix this by using EVT::getVectorVT instead.

In the added test, v7i8 is a SimpleVT, but v7i64 isn't.
@llvmbot
Copy link
Member

llvmbot commented Oct 31, 2025

@llvm/pr-subscribers-backend-risc-v

Author: Craig Topper (topperc)

Changes

This function doesn't work well when the type is a SimpleVT, but the changed type isn't. We need an LLVMContext to make an non-SimpleVT, but there's nowhere to get it from.

Fix this by using EVT::getVectorVT instead.

In the added test, v7i8 is a SimpleVT, but v7i64 isn't.

Full diff: https://github.com/llvm/llvm-project/pull/165829.diff

2 Files Affected:

  • (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+3-1)
  • (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll (+253)
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 56881f71934c4..c6a8b8481c94f 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -19794,7 +19794,9 @@ legalizeScatterGatherIndexType(SDLoc DL, SDValue &Index,
     // LLVM's legalization take care of the splitting.
     // FIXME: LLVM can't split VP_GATHER or VP_SCATTER yet.
     Index = DAG.getNode(ISD::SIGN_EXTEND, DL,
-                        IndexVT.changeVectorElementType(XLenVT), Index);
+                        EVT::getVectorVT(*DAG.getContext(), XLenVT,
+                                         IndexVT.getVectorElementCount()),
+                        Index);
   }
   IndexType = ISD::UNSIGNED_SCALED;
   return true;
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
index 4c35b2506d3e4..7e6f2c76e5881 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
@@ -15265,6 +15265,259 @@ define <4 x i32> @masked_gather_widen_sew_negative_stride(ptr %base) {
   ret <4 x i32> %x
 }
 
+define <7 x i8> @mgather_baseidx_v7i8(ptr %base, <7 x i8> %idxs, <7 x i1> %m, <7 x i8> %passthru) {
+; RV32-LABEL: mgather_baseidx_v7i8:
+; RV32:       # %bb.0:
+; RV32-NEXT:    li a1, 127
+; RV32-NEXT:    vsetivli zero, 8, e32, m2, ta, ma
+; RV32-NEXT:    vmv.s.x v10, a1
+; RV32-NEXT:    vmand.mm v0, v0, v10
+; RV32-NEXT:    vsext.vf4 v10, v8
+; RV32-NEXT:    vsetvli zero, zero, e8, mf2, ta, mu
+; RV32-NEXT:    vluxei32.v v9, (a0), v10, v0.t
+; RV32-NEXT:    vmv1r.v v8, v9
+; RV32-NEXT:    ret
+;
+; RV64V-LABEL: mgather_baseidx_v7i8:
+; RV64V:       # %bb.0:
+; RV64V-NEXT:    li a1, 127
+; RV64V-NEXT:    vsetivli zero, 8, e64, m4, ta, ma
+; RV64V-NEXT:    vmv.s.x v10, a1
+; RV64V-NEXT:    vmand.mm v0, v0, v10
+; RV64V-NEXT:    vsext.vf8 v12, v8
+; RV64V-NEXT:    vsetvli zero, zero, e8, mf2, ta, mu
+; RV64V-NEXT:    vluxei64.v v9, (a0), v12, v0.t
+; RV64V-NEXT:    vmv1r.v v8, v9
+; RV64V-NEXT:    ret
+;
+; RV64ZVE32F-LABEL: mgather_baseidx_v7i8:
+; RV64ZVE32F:       # %bb.0:
+; RV64ZVE32F-NEXT:    addi sp, sp, -16
+; RV64ZVE32F-NEXT:    .cfi_def_cfa_offset 16
+; RV64ZVE32F-NEXT:    .cfi_remember_state
+; RV64ZVE32F-NEXT:    li a1, 64
+; RV64ZVE32F-NEXT:    addi a2, sp, 8
+; RV64ZVE32F-NEXT:    vsetvli zero, a1, e8, m4, ta, ma
+; RV64ZVE32F-NEXT:    vsm.v v0, (a2)
+; RV64ZVE32F-NEXT:    ld a1, 8(sp)
+; RV64ZVE32F-NEXT:    andi a2, a1, 1
+; RV64ZVE32F-NEXT:    beqz a2, .LBB132_2
+; RV64ZVE32F-NEXT:  # %bb.1: # %cond.load
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v8
+; RV64ZVE32F-NEXT:    add a2, a0, a2
+; RV64ZVE32F-NEXT:    lbu a2, 0(a2)
+; RV64ZVE32F-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 1
+; RV64ZVE32F-NEXT:    vmv.v.x v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 2
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 3
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 4
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 5
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 6
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vslide1down.vx v9, v10, a2
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 1
+; RV64ZVE32F-NEXT:  .LBB132_2: # %else
+; RV64ZVE32F-NEXT:    andi a2, a1, 2
+; RV64ZVE32F-NEXT:    beqz a2, .LBB132_4
+; RV64ZVE32F-NEXT:  # %bb.3: # %cond.load1
+; RV64ZVE32F-NEXT:    vsetivli zero, 1, e8, mf4, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v8, 1
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 2
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v10
+; RV64ZVE32F-NEXT:    add a3, a0, a3
+; RV64ZVE32F-NEXT:    lbu a3, 0(a3)
+; RV64ZVE32F-NEXT:    vmv.v.x v10, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v11
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 3
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a3
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v11
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 4
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v11
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 5
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 6
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a3
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v11
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a2
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a3
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vslide1down.vx v9, v10, a2
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 1
+; RV64ZVE32F-NEXT:  .LBB132_4: # %else2
+; RV64ZVE32F-NEXT:    andi a2, a1, 4
+; RV64ZVE32F-NEXT:    vsetivli zero, 2, e8, mf4, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v8, 2
+; RV64ZVE32F-NEXT:    beqz a2, .LBB132_6
+; RV64ZVE32F-NEXT:  # %bb.5: # %cond.load4
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 1
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v9
+; RV64ZVE32F-NEXT:    vslidedown.vi v12, v9, 3
+; RV64ZVE32F-NEXT:    vmv.x.s a4, v11
+; RV64ZVE32F-NEXT:    vmv.v.x v11, a3
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v12
+; RV64ZVE32F-NEXT:    vslidedown.vi v12, v9, 4
+; RV64ZVE32F-NEXT:    add a2, a0, a2
+; RV64ZVE32F-NEXT:    lbu a2, 0(a2)
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a4
+; RV64ZVE32F-NEXT:    vmv.x.s a4, v12
+; RV64ZVE32F-NEXT:    vslidedown.vi v12, v9, 5
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 6
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v12
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a3
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a4
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vslide1down.vx v9, v11, a2
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 1
+; RV64ZVE32F-NEXT:  .LBB132_6: # %else5
+; RV64ZVE32F-NEXT:    andi a2, a1, 8
+; RV64ZVE32F-NEXT:    beqz a2, .LBB132_8
+; RV64ZVE32F-NEXT:  # %bb.7: # %cond.load7
+; RV64ZVE32F-NEXT:    vsetivli zero, 1, e8, mf4, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v10, 1
+; RV64ZVE32F-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 1
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 2
+; RV64ZVE32F-NEXT:    vmv.x.s a4, v11
+; RV64ZVE32F-NEXT:    vmv.v.x v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 4
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a4
+; RV64ZVE32F-NEXT:    vmv.x.s a4, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 5
+; RV64ZVE32F-NEXT:    add a3, a0, a3
+; RV64ZVE32F-NEXT:    lbu a3, 0(a3)
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 6
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v11, a3
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a4
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vslide1down.vx v9, v10, a2
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 1
+; RV64ZVE32F-NEXT:  .LBB132_8: # %else8
+; RV64ZVE32F-NEXT:    andi a2, a1, 16
+; RV64ZVE32F-NEXT:    vsetivli zero, 4, e8, mf2, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v8, v8, 4
+; RV64ZVE32F-NEXT:    bnez a2, .LBB132_13
+; RV64ZVE32F-NEXT:  # %bb.9: # %else11
+; RV64ZVE32F-NEXT:    andi a2, a1, 32
+; RV64ZVE32F-NEXT:    bnez a2, .LBB132_14
+; RV64ZVE32F-NEXT:  .LBB132_10: # %else14
+; RV64ZVE32F-NEXT:    andi a1, a1, 64
+; RV64ZVE32F-NEXT:    beqz a1, .LBB132_12
+; RV64ZVE32F-NEXT:  .LBB132_11: # %cond.load16
+; RV64ZVE32F-NEXT:    vsetivli zero, 2, e8, mf4, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v8, v8, 2
+; RV64ZVE32F-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 1
+; RV64ZVE32F-NEXT:    vmv.x.s a1, v9
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v8
+; RV64ZVE32F-NEXT:    vmv.v.x v8, a1
+; RV64ZVE32F-NEXT:    vmv.x.s a1, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 2
+; RV64ZVE32F-NEXT:    add a0, a0, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 3
+; RV64ZVE32F-NEXT:    vslide1down.vx v8, v8, a1
+; RV64ZVE32F-NEXT:    vmv.x.s a1, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 4
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 5
+; RV64ZVE32F-NEXT:    vslide1down.vx v8, v8, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    lbu a0, 0(a0)
+; RV64ZVE32F-NEXT:    vslide1down.vx v8, v8, a1
+; RV64ZVE32F-NEXT:    vslide1down.vx v8, v8, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a1, v9
+; RV64ZVE32F-NEXT:    vslide1down.vx v8, v8, a1
+; RV64ZVE32F-NEXT:    vslide1down.vx v8, v8, a0
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v8, 1
+; RV64ZVE32F-NEXT:  .LBB132_12: # %else17
+; RV64ZVE32F-NEXT:    vsetivli zero, 1, e8, m1, ta, ma
+; RV64ZVE32F-NEXT:    vmv1r.v v8, v9
+; RV64ZVE32F-NEXT:    addi sp, sp, 16
+; RV64ZVE32F-NEXT:    .cfi_def_cfa_offset 0
+; RV64ZVE32F-NEXT:    ret
+; RV64ZVE32F-NEXT:  .LBB132_13: # %cond.load10
+; RV64ZVE32F-NEXT:    .cfi_restore_state
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v8
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 1
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v9
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 2
+; RV64ZVE32F-NEXT:    vmv.x.s a4, v10
+; RV64ZVE32F-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
+; RV64ZVE32F-NEXT:    vmv.v.x v10, a3
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v11
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 3
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a4
+; RV64ZVE32F-NEXT:    vmv.x.s a4, v11
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 5
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 6
+; RV64ZVE32F-NEXT:    add a2, a0, a2
+; RV64ZVE32F-NEXT:    lbu a2, 0(a2)
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a3
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v11
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a4
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a2
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a3
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vslide1down.vx v9, v10, a2
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 1
+; RV64ZVE32F-NEXT:    andi a2, a1, 32
+; RV64ZVE32F-NEXT:    beqz a2, .LBB132_10
+; RV64ZVE32F-NEXT:  .LBB132_14: # %cond.load13
+; RV64ZVE32F-NEXT:    vsetivli zero, 1, e8, mf4, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v8, 1
+; RV64ZVE32F-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
+; RV64ZVE32F-NEXT:    vslidedown.vi v11, v9, 1
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vmv.x.s a3, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 2
+; RV64ZVE32F-NEXT:    vmv.x.s a4, v11
+; RV64ZVE32F-NEXT:    vmv.v.x v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 3
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a4
+; RV64ZVE32F-NEXT:    vmv.x.s a4, v10
+; RV64ZVE32F-NEXT:    vslidedown.vi v10, v9, 4
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 6
+; RV64ZVE32F-NEXT:    add a3, a0, a3
+; RV64ZVE32F-NEXT:    lbu a3, 0(a3)
+; RV64ZVE32F-NEXT:    vslide1down.vx v11, v11, a2
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v10
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v11, a4
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a2
+; RV64ZVE32F-NEXT:    vslide1down.vx v10, v10, a3
+; RV64ZVE32F-NEXT:    vmv.x.s a2, v9
+; RV64ZVE32F-NEXT:    vslide1down.vx v9, v10, a2
+; RV64ZVE32F-NEXT:    vslidedown.vi v9, v9, 1
+; RV64ZVE32F-NEXT:    andi a1, a1, 64
+; RV64ZVE32F-NEXT:    bnez a1, .LBB132_11
+; RV64ZVE32F-NEXT:    j .LBB132_12
+  %ptrs = getelementptr inbounds i8, ptr %base, <7 x i8> %idxs
+  %v = call <7 x i8> @llvm.masked.gather.v7i8.v7p0(<7 x ptr> %ptrs, i32 1, <7 x i1> %m, <7 x i8> %passthru)
+  ret <7 x i8> %v
+}
+
 ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
 ; RV32V-ZVFH: {{.*}}
 ; RV32V-ZVFHMIN: {{.*}}

lukel97
lukel97 approved these changes Oct 31, 2025
View reviewed changes
Copy link
Contributor

@lukel97 lukel97 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@topperc topperc merged commit e5f3d7f into llvm:main Oct 31, 2025
11 of 12 checks passed
@topperc topperc deleted the pr/changeVectorElementType branch October 31, 2025 17:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@preames preames preames approved these changes

@lukel97 lukel97 lukel97 approved these changes

@wangpc-pp wangpc-pp Awaiting requested review from wangpc-pp

Assignees

No one assigned

Labels

backend:RISC-V

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@topperc @llvmbot @preames @lukel97