[llvm] 3e678cb - [RISCV] Don't emit fractional VIDs with negative steps

[Fraser Cormack via llvm-commits]

Author: Fraser Cormack
Date: 2022-04-21T07:00:34+01:00
New Revision: 3e678cb77264907fbc2899c291ce23af308073ff

URL: https://github.com/llvm/llvm-project/commit/3e678cb77264907fbc2899c291ce23af308073ff
DIFF: https://github.com/llvm/llvm-project/commit/3e678cb77264907fbc2899c291ce23af308073ff.diff

LOG: [RISCV] Don't emit fractional VIDs with negative steps

We can't shift-right negative numbers to divide them, so avoid emitting
such sequences. Use negative numerators as a proxy for this situation, since
the indices are always non-negative.

An alternative strategy could be to add a compiler flag to emit division
instructions, which would at least allow us to test the VID sequence
matching itself.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D123796

Added: 
    

Modified: 
    llvm/lib/Target/RISCV/RISCVISelLowering.cpp
    llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 2d64acdaee128..6e19b831b4ac8 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -2186,7 +2186,8 @@ static SDValue lowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
     // a single addi instruction.
     if (((StepOpcode == ISD::MUL && isInt<12>(SplatStepVal)) ||
          (StepOpcode == ISD::SHL && isUInt<5>(SplatStepVal))) &&
-        isPowerOf2_32(StepDenominator) && isInt<5>(Addend)) {
+        isPowerOf2_32(StepDenominator) &&
+        (SplatStepVal >= 0 || StepDenominator == 1) && isInt<5>(Addend)) {
       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

diff  --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll
index 925ea684b5934..8c9d056b0bda0 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll
@@ -746,17 +746,17 @@ define <4 x i8> @buildvec_not_vid_v4i8_2() {
   ret <4 x i8> <i8 3, i8 3, i8 1, i8 0>
 }
 
-; FIXME: This is not a valid way to emit this vid sequence: shift-right for
-; division only works for non-negative numbers!
+; We match this as a VID sequence (-3 / 8) + 5 but choose not to introduce
+; division to compute it.
 define <16 x i8> @buildvec_not_vid_v16i8() {
 ; CHECK-LABEL: buildvec_not_vid_v16i8:
 ; CHECK:       # %bb.0:
+; CHECK-NEXT:    li a0, 3
 ; CHECK-NEXT:    vsetivli zero, 16, e8, m1, ta, mu
-; CHECK-NEXT:    vid.v v8
-; CHECK-NEXT:    li a0, -3
-; CHECK-NEXT:    vmul.vx v8, v8, a0
-; CHECK-NEXT:    vsrl.vi v8, v8, 3
-; CHECK-NEXT:    vadd.vi v8, v8, 5
+; CHECK-NEXT:    vmv.s.x v9, a0
+; CHECK-NEXT:    vmv.v.i v8, 0
+; CHECK-NEXT:    vsetivli zero, 7, e8, m1, tu, mu
+; CHECK-NEXT:    vslideup.vi v8, v9, 6
 ; CHECK-NEXT:    ret
   ret <16 x i8> <i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 3, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 0, i8 0>
 }