[RISCV] Allow undef elements in isDeinterleaveShuffle

[preames]

This allows us to form vnsrl deinterleaves from non-power-of-two shuffles after they've been legalized to a power of two.

[llvmbot]

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

Author: Philip Reames (preames)

Changes

This allows us to form vnsrl deinterleaves from non-power-of-two shuffles after they've been legalized to a power of two.

---

Full diff: https://github.com/llvm/llvm-project/pull/114585.diff

3 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+3-3)
• (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-interleaved-access.ll (+7-33)
• (modified) llvm/test/CodeGen/RISCV/rvv/vector-deinterleave-fixed.ll (-3)

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 920b06c7ba6ecd..047f74f3e6db9f 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -4475,10 +4475,10 @@ static bool isDeinterleaveShuffle(MVT VT, MVT ContainerVT, SDValue V1,
   if (Mask[0] != 0 && Mask[0] != 1)
     return false;
 
-  // The others must increase by 2 each time.
-  // TODO: Support undef elements?
+  // The others must increase by 2 each time (or be undef).
   for (unsigned i = 1; i != Mask.size(); ++i)
-    if (Mask[i] != Mask[i - 1] + 2)
+    if (Mask[i] != Mask[i - 1] + 2 &&
+        Mask[i] != -1)
       return false;
 
   return true;
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-interleaved-access.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-interleaved-access.ll
index 5911e8248f2995..b3cc834c70a1cc 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-interleaved-access.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-interleaved-access.ll
@@ -12,46 +12,20 @@ define {<3 x i32>, <3 x i32>} @load_factor2_v3(ptr %ptr) {
 ; RV32:       # %bb.0:
 ; RV32-NEXT:    vsetivli zero, 6, e32, m2, ta, ma
 ; RV32-NEXT:    vle32.v v10, (a0)
-; RV32-NEXT:    vsetivli zero, 2, e32, m1, ta, ma
-; RV32-NEXT:    vslidedown.vi v9, v10, 2
-; RV32-NEXT:    vsetivli zero, 2, e32, mf2, ta, ma
-; RV32-NEXT:    vwaddu.vv v8, v10, v9
-; RV32-NEXT:    li a0, -1
-; RV32-NEXT:    vwmaccu.vx v8, a0, v9
-; RV32-NEXT:    vmv.v.i v0, 4
-; RV32-NEXT:    vsetivli zero, 4, e32, m2, ta, ma
-; RV32-NEXT:    vslidedown.vi v12, v10, 4
-; RV32-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; RV32-NEXT:    vrgather.vi v8, v12, 0, v0.t
-; RV32-NEXT:    vid.v v9
-; RV32-NEXT:    vadd.vv v9, v9, v9
-; RV32-NEXT:    vadd.vi v11, v9, 1
-; RV32-NEXT:    vrgather.vv v9, v10, v11
-; RV32-NEXT:    vrgather.vi v9, v12, 1, v0.t
+; RV32-NEXT:    vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT:    vnsrl.wi v8, v10, 0
+; RV32-NEXT:    li a0, 32
+; RV32-NEXT:    vnsrl.wx v9, v10, a0
 ; RV32-NEXT:    ret
 ;
 ; RV64-LABEL: load_factor2_v3:
 ; RV64:       # %bb.0:
 ; RV64-NEXT:    vsetivli zero, 6, e32, m2, ta, ma
 ; RV64-NEXT:    vle32.v v10, (a0)
+; RV64-NEXT:    li a0, 32
 ; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, ma
-; RV64-NEXT:    vid.v v8
-; RV64-NEXT:    vadd.vv v8, v8, v8
-; RV64-NEXT:    vadd.vi v8, v8, 1
-; RV64-NEXT:    vrgather.vv v9, v10, v8
-; RV64-NEXT:    vmv.v.i v0, 4
-; RV64-NEXT:    vsetivli zero, 4, e32, m2, ta, ma
-; RV64-NEXT:    vslidedown.vi v12, v10, 4
-; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; RV64-NEXT:    vrgather.vi v9, v12, 1, v0.t
-; RV64-NEXT:    vsetivli zero, 2, e32, m1, ta, ma
-; RV64-NEXT:    vslidedown.vi v11, v10, 2
-; RV64-NEXT:    vsetivli zero, 2, e32, mf2, ta, ma
-; RV64-NEXT:    vwaddu.vv v8, v10, v11
-; RV64-NEXT:    li a0, -1
-; RV64-NEXT:    vwmaccu.vx v8, a0, v11
-; RV64-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; RV64-NEXT:    vrgather.vi v8, v12, 0, v0.t
+; RV64-NEXT:    vnsrl.wx v9, v10, a0
+; RV64-NEXT:    vnsrl.wi v8, v10, 0
 ; RV64-NEXT:    ret
   %interleaved.vec = load <6 x i32>, ptr %ptr
   %v0 = shufflevector <6 x i32> %interleaved.vec, <6 x i32> poison, <3 x i32> <i32 0, i32 2, i32 4>
diff --git a/llvm/test/CodeGen/RISCV/rvv/vector-deinterleave-fixed.ll b/llvm/test/CodeGen/RISCV/rvv/vector-deinterleave-fixed.ll
index bc32518b671953..b2973826d65ded 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vector-deinterleave-fixed.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vector-deinterleave-fixed.ll
@@ -269,6 +269,3 @@ declare {<2 x float>, <2 x float>} @llvm.vector.deinterleave2.v4f32(<4 x float>)
 declare {<8 x half>, <8 x half>} @llvm.vector.deinterleave2.v16f16(<16 x half>)
 declare {<4 x float>, <4 x float>} @llvm.vector.deinterleave2.v8f32(<8 x float>)
 declare {<2 x double>, <2 x double>} @llvm.vector.deinterleave2.v4f64(<4 x double>)
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; RV32: {{.*}}
-; RV64: {{.*}}

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[preames]

There's a bug here - specifically, this allows 1 in an arbitrary position following an out of bounds index. Will refresh with a fixed version on Monday.

[topperc]

LGTM