[RISCV] Use getSplatBuildVector instead of getSplatVector for fixed vectors.

[craig.topper]

Summary

The splat_vector will be legalized to build_vector eventually
anyway. This patch makes it take fewer steps.

Unfortunately, this results in some codegen changes. It looks
like it comes down to how the nodes were ordered in the topological
sort for isel. Because the build_vector is created earlier we end up
with a different ordering of nodes.

[frasercrmck]

Seems sensible to me, thanks.

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll

571

Huh, this is the only line that differs between RV32 and RV64? Not sure we've seen something like that before, have we?

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

 //===-- RISCVISelLowering.cpp - RISCV DAG Lowering Implementation  --------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
if (((StepOpcode == ISD::MUL && isInt<12>(SplatStepVal)) ||
       SDValue VID = DAG.getNode(RISCVISD::VID_VL, DL, ContainerVT, Mask, VL);
       // Convert right out of the scalable type so we can use standard ISD
       // nodes for the rest of the computation. If we used scalable types with
       // these, we'd lose the fixed-length vector info and generate worse
       // vsetvli code.
       VID = convertFromScalableVector(VT, VID, DAG, Subtarget);
       if ((StepOpcode == ISD::MUL && SplatStepVal != 1) ||
           (StepOpcode == ISD::SHL && SplatStepVal != 0)) {
-        SDValue SplatStep = DAG.getSplatVector(
+        SDValue SplatStep = DAG.getSplatBuildVector(
             VT, DL, DAG.getConstant(SplatStepVal, DL, XLenVT));
         VID = DAG.getNode(StepOpcode, DL, VT, VID, SplatStep);
       }
       if (StepDenominator != 1) {
-        SDValue SplatStep = DAG.getSplatVector(
+        SDValue SplatStep = DAG.getSplatBuildVector(
             VT, DL, DAG.getConstant(Log2_64(StepDenominator), DL, XLenVT));
         VID = DAG.getNode(ISD::SRL, DL, VT, VID, SplatStep);
       }
       if (Addend != 0 || Negate) {
-        SDValue SplatAddend =
-            DAG.getSplatVector(VT, DL, DAG.getConstant(Addend, DL, XLenVT));
+        SDValue SplatAddend = DAG.getSplatBuildVector(
+            VT, DL, DAG.getConstant(Addend, DL, XLenVT));
         VID = DAG.getNode(Negate ? ISD::SUB : ISD::ADD, DL, VT, SplatAddend, VID);
       }
       return VID;
     }
   }
 
   // Attempt to detect "hidden" splats, which only reveal themselves as splats
   // when re-interpreted as a vector with a larger element type. For example,

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll

 ; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
 ; CHECK-NEXT:    vse16.v v8, (a0)
 ; CHECK-NEXT:    ret
   store <4 x i16> <i16 -127, i16 -1, i16 -127, i16 -1>, <4 x i16>* %x
   ret void
 }
 
 define void @buildvec_vid_step1o2_v4i32(<4 x i32>* %z0, <4 x i32>* %z1, <4 x i32>* %z2, <4 x i32>* %z3, <4 x i32>* %z4, <4 x i32>* %z5, <4 x i32>* %z6) {
-; CHECK-LABEL: buildvec_vid_step1o2_v4i32:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vid.v v8
-; CHECK-NEXT:    vsrl.vi v8, v8, 1
-; CHECK-NEXT:    vse32.v v8, (a0)
-; CHECK-NEXT:    vse32.v v8, (a1)
-; CHECK-NEXT:    vse32.v v8, (a2)
-; CHECK-NEXT:    vse32.v v8, (a3)
-; CHECK-NEXT:    vse32.v v8, (a4)
-; CHECK-NEXT:    vmv.s.x v8, zero
-; CHECK-NEXT:    vmv.v.i v9, 1
-; CHECK-NEXT:    vsetivli zero, 2, e32, m1, tu, mu
-; CHECK-NEXT:    vslideup.vi v9, v8, 1
-; CHECK-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; CHECK-NEXT:    vse32.v v9, (a5)
-; CHECK-NEXT:    li a0, 1
-; CHECK-NEXT:    vmv.s.x v8, a0
-; CHECK-NEXT:    vmv.v.i v9, 0
-; CHECK-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
-; CHECK-NEXT:    vslideup.vi v9, v8, 3
-; CHECK-NEXT:    vse32.v v9, (a6)
-; CHECK-NEXT:    ret
+; RV32-LABEL: buildvec_vid_step1o2_v4i32:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV32-NEXT:    vid.v v8
+; RV32-NEXT:    vsrl.vi v8, v8, 1
+; RV32-NEXT:    vse32.v v8, (a0)
+; RV32-NEXT:    vse32.v v8, (a1)
+; RV32-NEXT:    vmv.v.i v9, 1
+; RV32-NEXT:    vse32.v v8, (a2)
+; RV32-NEXT:    vse32.v v8, (a3)
+; RV32-NEXT:    vse32.v v8, (a4)
+; RV32-NEXT:    vmv.s.x v8, zero
+; RV32-NEXT:    vsetivli zero, 2, e32, m1, tu, mu
+; RV32-NEXT:    vslideup.vi v9, v8, 1
+; RV32-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV32-NEXT:    vse32.v v9, (a5)
+; RV32-NEXT:    li a0, 1
+; RV32-NEXT:    vmv.s.x v8, a0
+; RV32-NEXT:    vmv.v.i v9, 0
+; RV32-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
+; RV32-NEXT:    vslideup.vi v9, v8, 3
+; RV32-NEXT:    vse32.v v9, (a6)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: buildvec_vid_step1o2_v4i32:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV64-NEXT:    vid.v v8
+; RV64-NEXT:    vsrl.vi v8, v8, 1
+; RV64-NEXT:    vse32.v v8, (a0)
+; RV64-NEXT:    vmv.v.i v9, 1
+; RV64-NEXT:    vse32.v v8, (a1)
+; RV64-NEXT:    vse32.v v8, (a2)
+; RV64-NEXT:    vse32.v v8, (a3)
+; RV64-NEXT:    vse32.v v8, (a4)
+; RV64-NEXT:    vmv.s.x v8, zero
+; RV64-NEXT:    vsetivli zero, 2, e32, m1, tu, mu
+; RV64-NEXT:    vslideup.vi v9, v8, 1
+; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
+; RV64-NEXT:    vse32.v v9, (a5)
+; RV64-NEXT:    li a0, 1
+; RV64-NEXT:    vmv.s.x v8, a0
+; RV64-NEXT:    vmv.v.i v9, 0
+; RV64-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
+; RV64-NEXT:    vslideup.vi v9, v8, 3
+; RV64-NEXT:    vse32.v v9, (a6)
+; RV64-NEXT:    ret
   store <4 x i32> <i32 0, i32 0, i32 1, i32 1>, <4 x i32>* %z0
   store <4 x i32> <i32 0, i32 0, i32 1, i32 undef>, <4 x i32>* %z1
   store <4 x i32> <i32 0, i32 undef, i32 1, i32 1>, <4 x i32>* %z2
   store <4 x i32> <i32 undef, i32 0, i32 undef, i32 1>, <4 x i32>* %z3
   store <4 x i32> <i32 0, i32 undef, i32 1, i32 undef>, <4 x i32>* %z4
   ; We don't catch this one
   store <4 x i32> <i32 undef, i32 0, i32 1, i32 1>, <4 x i32>* %z5
   ; We catch this one but as VID/3 rather than VID/2
   store <4 x i32> <i32 0, i32 0, i32 undef, i32 1>, <4 x i32>* %z6
   ret void
 }
 
 define void @buildvec_vid_step1o2_add3_v4i16(<4 x i16>* %z0, <4 x i16>* %z1, <4 x i16>* %z2, <4 x i16>* %z3, <4 x i16>* %z4, <4 x i16>* %z5, <4 x i16>* %z6) {
 ; CHECK-LABEL: buildvec_vid_step1o2_add3_v4i16:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
 ; CHECK-NEXT:    vid.v v8
 ; CHECK-NEXT:    vsrl.vi v8, v8, 1
 ; CHECK-NEXT:    vadd.vi v8, v8, 3
 ; CHECK-NEXT:    vse16.v v8, (a0)
+; CHECK-NEXT:    vmv.v.i v9, 3
 ; CHECK-NEXT:    vse16.v v8, (a1)
 ; CHECK-NEXT:    vse16.v v8, (a2)
 ; CHECK-NEXT:    vse16.v v8, (a3)
 ; CHECK-NEXT:    vse16.v v8, (a4)
 ; CHECK-NEXT:    li a0, 3
 ; CHECK-NEXT:    vmv.s.x v8, a0
-; CHECK-NEXT:    vmv.v.i v9, 4
+; CHECK-NEXT:    vmv.v.i v10, 4
 ; CHECK-NEXT:    vsetivli zero, 2, e16, mf2, tu, mu
-; CHECK-NEXT:    vslideup.vi v9, v8, 1
+; CHECK-NEXT:    vslideup.vi v10, v8, 1
 ; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
-; CHECK-NEXT:    vse16.v v9, (a5)
+; CHECK-NEXT:    vse16.v v10, (a5)
 ; CHECK-NEXT:    li a0, 4
 ; CHECK-NEXT:    vmv.s.x v8, a0
-; CHECK-NEXT:    vmv.v.i v9, 3
 ; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, tu, mu
 ; CHECK-NEXT:    vslideup.vi v9, v8, 3
 ; CHECK-NEXT:    vse16.v v9, (a6)
 ; CHECK-NEXT:    ret
   store <4 x i16> <i16 3, i16 3, i16 4, i16 4>, <4 x i16>* %z0
   store <4 x i16> <i16 3, i16 3, i16 4, i16 undef>, <4 x i16>* %z1
   store <4 x i16> <i16 3, i16 undef, i16 4, i16 4>, <4 x i16>* %z2
   store <4 x i16> <i16 undef, i16 3, i16 undef, i16 4>, <4 x i16>* %z3