[RISCV] Change how MMO is rebuilt in lowerFixedLengthVectorLoadToRVV/lowerFixedLengthVectorStoreToRVV

[topperc]

Copy the pointer info, flags, alignment, AAInfo, and ranges, but let getLoad rebuild the MMO using the scalable type used for the the new load/store. This makes sure the LLT minimum size matches the ContainerVT minimum size. This is important since vscale_range may have been used to determine that the fixed vector was the exact size of a scalable vector.

Fixes #88799

[llvmbot]

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

Author: Craig Topper (topperc)

Changes

Copy the pointer info, flags, alignment, AAInfo, and ranges, but let getLoad rebuild the MMO using the scalable type used for the the new load/store. This makes sure the LLT minimum size matches the ContainerVT minimum size. This is important since vscale_range may have been used to determine that the fixed vector was the exact size of a scalable vector.

Fixes #88799

---

Full diff: https://github.com/llvm/llvm-project/pull/88811.diff

2 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+5-14)
• (added) llvm/test/CodeGen/RISCV/rvv/pr88799.ll (+19)

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 27387595164a46..76a351ada67182 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -10403,14 +10403,10 @@ RISCVTargetLowering::lowerFixedLengthVectorLoadToRVV(SDValue Op,
   if (MinVLMAX == MaxVLMAX && MinVLMAX == VT.getVectorNumElements() &&
       getLMUL1VT(ContainerVT).bitsLE(ContainerVT)) {
     MachineMemOperand *MMO = Load->getMemOperand();
-    MachineFunction &MF = DAG.getMachineFunction();
-    MMO = MF.getMachineMemOperand(
-        MMO, MMO->getPointerInfo(),
-        MMO->getMemoryType().isValid()
-            ? LLT::scalable_vector(1, MMO->getMemoryType().getSizeInBits())
-            : MMO->getMemoryType());
     SDValue NewLoad =
-        DAG.getLoad(ContainerVT, DL, Load->getChain(), Load->getBasePtr(), MMO);
+        DAG.getLoad(ContainerVT, DL, Load->getChain(), Load->getBasePtr(),
+                    MMO->getPointerInfo(), MMO->getBaseAlign(), MMO->getFlags(),
+                    MMO->getAAInfo(), MMO->getRanges());
     SDValue Result = convertFromScalableVector(VT, NewLoad, DAG, Subtarget);
     return DAG.getMergeValues({Result, NewLoad.getValue(1)}, DL);
   }
@@ -10470,14 +10466,9 @@ RISCVTargetLowering::lowerFixedLengthVectorStoreToRVV(SDValue Op,
   if (MinVLMAX == MaxVLMAX && MinVLMAX == VT.getVectorNumElements() &&
       getLMUL1VT(ContainerVT).bitsLE(ContainerVT)) {
     MachineMemOperand *MMO = Store->getMemOperand();
-    MachineFunction &MF = DAG.getMachineFunction();
-    MMO = MF.getMachineMemOperand(
-        MMO, MMO->getPointerInfo(),
-        MMO->getMemoryType().isValid()
-            ? LLT::scalable_vector(1, MMO->getMemoryType().getSizeInBits())
-            : MMO->getMemoryType());
     return DAG.getStore(Store->getChain(), DL, NewValue, Store->getBasePtr(),
-                        MMO);
+                        MMO->getPointerInfo(), MMO->getBaseAlign(),
+                        MMO->getFlags(), MMO->getAAInfo());
   }
 
   SDValue VL = getVLOp(VT.getVectorNumElements(), ContainerVT, DL, DAG,
diff --git a/llvm/test/CodeGen/RISCV/rvv/pr88799.ll b/llvm/test/CodeGen/RISCV/rvv/pr88799.ll
new file mode 100644
index 00000000000000..7212a789f9e7e0
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/pr88799.ll
@@ -0,0 +1,19 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc < %s -mtriple=riscv64-unknown-linux-gnu -mattr=+v | FileCheck %s
+
+define i32 @main() vscale_range(2,2) {
+; CHECK-LABEL: main:
+; CHECK:       # %bb.0: # %vector.body
+; CHECK-NEXT:    lui a0, 1040368
+; CHECK-NEXT:    addiw a0, a0, -144
+; CHECK-NEXT:    vl2re16.v v8, (a0)
+; CHECK-NEXT:    vs2r.v v8, (zero)
+; CHECK-NEXT:    li a0, 0
+; CHECK-NEXT:    ret
+vector.body:
+  %0 = load <16 x i16>, ptr getelementptr ([3 x [23 x [23 x i16]]], ptr null, i64 -10593, i64 1, i64 22, i64 0), align 16
+  store <16 x i16> %0, ptr null, align 2
+  %wide.load = load <vscale x 8 x i16>, ptr getelementptr ([3 x [23 x [23 x i16]]], ptr null, i64 -10593, i64 1, i64 22, i64 0), align 16
+  store <vscale x 8 x i16> %wide.load, ptr null, align 2
+  ret i32 0
+}

[hiraditya]

LGTM but I'd let original author/code-owner review it. Thanks for the quick fix!

[wangpc-pp]

LGTM.

[davemgreen]

Hi. Sorry for the bug and thanks for the fix. I think I missed that the typesize needed to be adjusted by the vscale range. The new code sounds good if its producing the right mmo sizes.