[WebAssembly] Expand operations not supported by SIMD

[tlively]

Summary

This prevents crashes in instruction selection when these operations
are used. The tests check that the scalar version of the instruction
is used where applicable, although some expansions do not use the
scalar version.

[aheejin]

The tests check that the scalar version of the instruction

is used where applicable, although some expansions do not use the
scalar version.

By skimming the tests, all are using their respective scalar versions.. which one does not use its scalar version and what does it use instead? Maybe worth one-line comment for that test (I haven't found which one)

[tlively]

• Add comments pointing out expansions that don't use scalar equivalents

[aheejin]

Oh sorry, I thought they used other SIMD instructions, which was the reason I suggested comments... I guess they are not very necessary after all. 'Expand' means not using SIMD. Sorry :(

[tlively]

• Bug fixes

llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp

if (Subtarget->hasSIMD128()) {
     for (auto Op : {ISD::VSELECT, ISD::SELECT_CC, ISD::SELECT}) {
       for (auto T : {MVT::v16i8, MVT::v8i16, MVT::v4i32, MVT::v4f32})
         setOperationAction(Op, T, Expand);
       if (Subtarget->hasUnimplementedSIMD128())
         for (auto T : {MVT::v2i64, MVT::v2f64})
           setOperationAction(Op, T, Expand);
     }
 
+    // Expand integer operations supported for scalars but not SIMD
+    for (auto Op : {ISD::CTLZ, ISD::CTTZ, ISD::CTPOP, ISD::SDIV, ISD::UDIV,
+                    ISD::SREM, ISD::UREM, ISD::ROTL, ISD::ROTR}) {
+      for (auto T : {MVT::v16i8, MVT::v8i16, MVT::v4i32})
+        setOperationAction(Op, T, Expand);
+      if (Subtarget->hasUnimplementedSIMD128())
+        setOperationAction(Op, MVT::v2i64, Expand);
+    }
+
+    // Expand float operations supported for scalars but not SIMD
+    for (auto Op : {ISD::FCEIL, ISD::FFLOOR, ISD::FTRUNC, ISD::FNEARBYINT,
+                    ISD::FCOPYSIGN}) {
+      setOperationAction(Op, MVT::v4f32, Expand);
+      if (Subtarget->hasUnimplementedSIMD128())
+        setOperationAction(Op, MVT::v2f64, Expand);
+    }
+
     // Expand additional SIMD ops that V8 hasn't implemented yet
     if (!Subtarget->hasUnimplementedSIMD128()) {
       setOperationAction(ISD::FSQRT, MVT::v4f32, Expand);
       setOperationAction(ISD::FDIV, MVT::v4f32, Expand);
     }
   }
 
   // As a special case, these operators use the type to mean the type to

llvm/test/CodeGen/WebAssembly/simd-unsupported.ll

+; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+unimplemented-simd128 | FileCheck %s
+
+; Test that operations that are supported by MVP but not SIMD are
+; properly unrolled.
+
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; ==============================================================================
+; 16 x i8
+; ==============================================================================
+
+; CHECK-LABEL: ctlz_v16i8:
+; CHECK: i32.clz
+declare <16 x i8> @llvm.ctlz.v16i8(<16 x i8>, i1)
+define <16 x i8> @ctlz_v16i8(<16 x i8> %x) {
+  %v = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %x, i1 false)
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: ctlz_v16i8_undef:
+; CHECK: i32.clz
+define <16 x i8> @ctlz_v16i8_undef(<16 x i8> %x) {
+  %v = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %x, i1 true)
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: cttz_v16i8:
+; CHECK: i32.ctz
+declare <16 x i8> @llvm.cttz.v16i8(<16 x i8>, i1)
+define <16 x i8> @cttz_v16i8(<16 x i8> %x) {
+  %v = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %x, i1 false)
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: cttz_v16i8_undef:
+; CHECK: i32.ctz
+define <16 x i8> @cttz_v16i8_undef(<16 x i8> %x) {
+  %v = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %x, i1 true)
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: ctpop_v16i8:
+; Note: expansion does not use i32.popcnt
+; CHECK: v128.and
+declare <16 x i8> @llvm.ctpop.v16i8(<16 x i8>)
+define <16 x i8> @ctpop_v16i8(<16 x i8> %x) {
+  %v = call <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %x)
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: sdiv_v16i8:
+; CHECK: i32.div_s
+define <16 x i8> @sdiv_v16i8(<16 x i8> %x, <16 x i8> %y) {
+  %v = sdiv <16 x i8> %x, %y
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: udiv_v16i8:
+; CHECK: i32.div_u
+define <16 x i8> @udiv_v16i8(<16 x i8> %x, <16 x i8> %y) {
+  %v = udiv <16 x i8> %x, %y
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: srem_v16i8:
+; CHECK: i32.rem_s
+define <16 x i8> @srem_v16i8(<16 x i8> %x, <16 x i8> %y) {
+  %v = srem <16 x i8> %x, %y
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: urem_v16i8:
+; CHECK: i32.rem_u
+define <16 x i8> @urem_v16i8(<16 x i8> %x, <16 x i8> %y) {
+  %v = urem <16 x i8> %x, %y
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: rotl_v16i8:
+; Note: expansion does not use i32.rotl
+; CHECK: i32.shl
+declare <16 x i8> @llvm.fshl.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)
+define <16 x i8> @rotl_v16i8(<16 x i8> %x, <16 x i8> %y) {
+  %v = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %x, <16 x i8> %x, <16 x i8> %y)
+  ret <16 x i8> %v
+}
+
+; CHECK-LABEL: rotr_v16i8:
+; Note: expansion does not use i32.rotr
+; CHECK: i32.shr_u
+declare <16 x i8> @llvm.fshr.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)
+define <16 x i8> @rotr_v16i8(<16 x i8> %x, <16 x i8> %y) {
+  %v = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %x, <16 x i8> %x, <16 x i8> %y)
+  ret <16 x i8> %v
+}
+
+; ==============================================================================
+; 8 x i16
+; ==============================================================================
+
+; CHECK-LABEL: ctlz_v8i16:
+; CHECK: i32.clz
+declare <8 x i16> @llvm.ctlz.v8i16(<8 x i16>, i1)
+define <8 x i16> @ctlz_v8i16(<8 x i16> %x) {
+  %v = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %x, i1 false)
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: ctlz_v8i16_undef:
+; CHECK: i32.clz
+define <8 x i16> @ctlz_v8i16_undef(<8 x i16> %x) {
+  %v = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %x, i1 true)
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: cttz_v8i16:
+; CHECK: i32.ctz
+declare <8 x i16> @llvm.cttz.v8i16(<8 x i16>, i1)
+define <8 x i16> @cttz_v8i16(<8 x i16> %x) {
+  %v = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %x, i1 false)
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: cttz_v8i16_undef:
+; CHECK: i32.ctz
+define <8 x i16> @cttz_v8i16_undef(<8 x i16> %x) {
+  %v = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %x, i1 true)
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: ctpop_v8i16:
+; Note: expansion does not use i32.popcnt
+; CHECK: v128.and
+declare <8 x i16> @llvm.ctpop.v8i16(<8 x i16>)
+define <8 x i16> @ctpop_v8i16(<8 x i16> %x) {
+  %v = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %x)
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: sdiv_v8i16:
+; CHECK: i32.div_s
+define <8 x i16> @sdiv_v8i16(<8 x i16> %x, <8 x i16> %y) {
+  %v = sdiv <8 x i16> %x, %y
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: udiv_v8i16:
+; CHECK: i32.div_u
+define <8 x i16> @udiv_v8i16(<8 x i16> %x, <8 x i16> %y) {
+  %v = udiv <8 x i16> %x, %y
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: srem_v8i16:
+; CHECK: i32.rem_s
+define <8 x i16> @srem_v8i16(<8 x i16> %x, <8 x i16> %y) {
+  %v = srem <8 x i16> %x, %y
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: urem_v8i16:
+; CHECK: i32.rem_u
+define <8 x i16> @urem_v8i16(<8 x i16> %x, <8 x i16> %y) {
+  %v = urem <8 x i16> %x, %y
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: rotl_v8i16:
+; Note: expansion does not use i32.rotl
+; CHECK: i32.shl
+declare <8 x i16> @llvm.fshl.v8i16(<8 x i16>, <8 x i16>, <8 x i16>)
+define <8 x i16> @rotl_v8i16(<8 x i16> %x, <8 x i16> %y) {
+  %v = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x, <8 x i16> %x, <8 x i16> %y)
+  ret <8 x i16> %v
+}
+
+; CHECK-LABEL: rotr_v8i16:
+; Note: expansion does not use i32.rotr
+; CHECK: i32.shr_u
+declare <8 x i16> @llvm.fshr.v8i16(<8 x i16>, <8 x i16>, <8 x i16>)
+define <8 x i16> @rotr_v8i16(<8 x i16> %x, <8 x i16> %y) {
+  %v = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x, <8 x i16> %x, <8 x i16> %y)
+  ret <8 x i16> %v
+}
+
+; ==============================================================================
+; 4 x i32
+; ==============================================================================
+
+; CHECK-LABEL: ctlz_v4i32:
+; CHECK: i32.clz
+declare <4 x i32> @llvm.ctlz.v4i32(<4 x i32>, i1)
+define <4 x i32> @ctlz_v4i32(<4 x i32> %x) {
+  %v = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %x, i1 false)
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: ctlz_v4i32_undef:
+; CHECK: i32.clz
+define <4 x i32> @ctlz_v4i32_undef(<4 x i32> %x) {
+  %v = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %x, i1 true)
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: cttz_v4i32:
+; CHECK: i32.ctz
+declare <4 x i32> @llvm.cttz.v4i32(<4 x i32>, i1)
+define <4 x i32> @cttz_v4i32(<4 x i32> %x) {
+  %v = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %x, i1 false)
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: cttz_v4i32_undef:
+; CHECK: i32.ctz
+define <4 x i32> @cttz_v4i32_undef(<4 x i32> %x) {
+  %v = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %x, i1 true)
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: ctpop_v4i32:
+; Note: expansion does not use i32.popcnt
+; CHECK: v128.and
+declare <4 x i32> @llvm.ctpop.v4i32(<4 x i32>)
+define <4 x i32> @ctpop_v4i32(<4 x i32> %x) {
+  %v = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %x)
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: sdiv_v4i32:
+; CHECK: i32.div_s
+define <4 x i32> @sdiv_v4i32(<4 x i32> %x, <4 x i32> %y) {
+  %v = sdiv <4 x i32> %x, %y
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: udiv_v4i32:
+; CHECK: i32.div_u
+define <4 x i32> @udiv_v4i32(<4 x i32> %x, <4 x i32> %y) {
+  %v = udiv <4 x i32> %x, %y
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: srem_v4i32:
+; CHECK: i32.rem_s
+define <4 x i32> @srem_v4i32(<4 x i32> %x, <4 x i32> %y) {
+  %v = srem <4 x i32> %x, %y
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: urem_v4i32:
+; CHECK: i32.rem_u
+define <4 x i32> @urem_v4i32(<4 x i32> %x, <4 x i32> %y) {
+  %v = urem <4 x i32> %x, %y
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: rotl_v4i32:
+; Note: expansion does not use i32.rotl
+; CHECK: i32.shl
+declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+define <4 x i32> @rotl_v4i32(<4 x i32> %x, <4 x i32> %y) {
+  %v = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %y)
+  ret <4 x i32> %v
+}
+
+; CHECK-LABEL: rotr_v4i32:
+; Note: expansion does not use i32.rotr
+; CHECK: i32.shr_u
+declare <4 x i32> @llvm.fshr.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+define <4 x i32> @rotr_v4i32(<4 x i32> %x, <4 x i32> %y) {
+  %v = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %y)
+  ret <4 x i32> %v
+}
+
+; ==============================================================================
+; 2 x i64
+; ==============================================================================
+
+; CHECK-LABEL: ctlz_v2i64:
+; CHECK: i64.clz
+declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1)
+define <2 x i64> @ctlz_v2i64(<2 x i64> %x) {
+  %v = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %x, i1 false)
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: ctlz_v2i64_undef:
+; CHECK: i64.clz
+define <2 x i64> @ctlz_v2i64_undef(<2 x i64> %x) {
+  %v = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %x, i1 true)
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: cttz_v2i64:
+; CHECK: i64.ctz
+declare <2 x i64> @llvm.cttz.v2i64(<2 x i64>, i1)
+define <2 x i64> @cttz_v2i64(<2 x i64> %x) {
+  %v = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %x, i1 false)
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: cttz_v2i64_undef:
+; CHECK: i64.ctz
+define <2 x i64> @cttz_v2i64_undef(<2 x i64> %x) {
+  %v = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %x, i1 true)
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: ctpop_v2i64:
+; CHECK: i64.popcnt
+declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>)
+define <2 x i64> @ctpop_v2i64(<2 x i64> %x) {
+  %v = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %x)
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: sdiv_v2i64:
+; CHECK: i64.div_s
+define <2 x i64> @sdiv_v2i64(<2 x i64> %x, <2 x i64> %y) {
+  %v = sdiv <2 x i64> %x, %y
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: udiv_v2i64:
+; CHECK: i64.div_u
+define <2 x i64> @udiv_v2i64(<2 x i64> %x, <2 x i64> %y) {
+  %v = udiv <2 x i64> %x, %y
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: srem_v2i64:
+; CHECK: i64.rem_s
+define <2 x i64> @srem_v2i64(<2 x i64> %x, <2 x i64> %y) {
+  %v = srem <2 x i64> %x, %y
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: urem_v2i64:
+; CHECK: i64.rem_u
+define <2 x i64> @urem_v2i64(<2 x i64> %x, <2 x i64> %y) {
+  %v = urem <2 x i64> %x, %y
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: rotl_v2i64:
+; Note: expansion does not use i64.rotl
+; CHECK: i64.shl
+declare <2 x i64> @llvm.fshl.v2i64(<2 x i64>, <2 x i64>, <2 x i64>)
+define <2 x i64> @rotl_v2i64(<2 x i64> %x, <2 x i64> %y) {
+  %v = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %x, <2 x i64> %y)
+  ret <2 x i64> %v
+}
+
+; CHECK-LABEL: rotr_v2i64:
+; Note: expansion does not use i64.rotr
+; CHECK: i64.shr_u
+declare <2 x i64> @llvm.fshr.v2i64(<2 x i64>, <2 x i64>, <2 x i64>)
+define <2 x i64> @rotr_v2i64(<2 x i64> %x, <2 x i64> %y) {
+  %v = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %x, <2 x i64> %y)
+  ret <2 x i64> %v
+}
+
+; ==============================================================================
+; 4 x f32
+; ==============================================================================
+
+; CHECK-LABEL: ceil_v4f32:
+; CHECK: f32.ceil
+declare <4 x float> @llvm.ceil.v4f32(<4 x float>)
+define <4 x float> @ceil_v4f32(<4 x float> %x) {
+  %v = call <4 x float> @llvm.ceil.v4f32(<4 x float> %x)
+  ret <4 x float> %v
+}
+
+; CHECK-LABEL: floor_v4f32:
+; CHECK: f32.floor
+declare <4 x float> @llvm.floor.v4f32(<4 x float>)
+define <4 x float> @floor_v4f32(<4 x float> %x) {
+  %v = call <4 x float> @llvm.floor.v4f32(<4 x float> %x)
+  ret <4 x float> %v
+}
+
+; CHECK-LABEL: trunc_v4f32:
+; CHECK: f32.trunc
+declare <4 x float> @llvm.trunc.v4f32(<4 x float>)
+define <4 x float> @trunc_v4f32(<4 x float> %x) {
+  %v = call <4 x float> @llvm.trunc.v4f32(<4 x float> %x)
+  ret <4 x float> %v
+}
+
+; CHECK-LABEL: nearbyint_v4f32:
+; CHECK: f32.nearest
+declare <4 x float> @llvm.nearbyint.v4f32(<4 x float>)
+define <4 x float> @nearbyint_v4f32(<4 x float> %x) {
+  %v = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %x)
+  ret <4 x float> %v
+}
+
+; CHECK-LABEL: copysign_v4f32:
+; CHECK: f32.copysign
+declare <4 x float> @llvm.copysign.v4f32(<4 x float>, <4 x float>)
+define <4 x float> @copysign_v4f32(<4 x float> %x, <4 x float> %y) {
+  %v = call <4 x float> @llvm.copysign.v4f32(<4 x float> %x, <4 x float> %y)
+  ret <4 x float> %v
+}
+
+; ==============================================================================
+; 2 x f64
+; ==============================================================================
+
+; CHECK-LABEL: ceil_v2f64:
+; CHECK: f64.ceil
+declare <2 x double> @llvm.ceil.v2f64(<2 x double>)
+define <2 x double> @ceil_v2f64(<2 x double> %x) {
+  %v = call <2 x double> @llvm.ceil.v2f64(<2 x double> %x)
+  ret <2 x double> %v
+}
+
+; CHECK-LABEL: floor_v2f64:
+; CHECK: f64.floor
+declare <2 x double> @llvm.floor.v2f64(<2 x double>)
+define <2 x double> @floor_v2f64(<2 x double> %x) {
+  %v = call <2 x double> @llvm.floor.v2f64(<2 x double> %x)
+  ret <2 x double> %v
+}
+
+; CHECK-LABEL: trunc_v2f64:
+; CHECK: f64.trunc
+declare <2 x double> @llvm.trunc.v2f64(<2 x double>)
+define <2 x double> @trunc_v2f64(<2 x double> %x) {
+  %v = call <2 x double> @llvm.trunc.v2f64(<2 x double> %x)
+  ret <2 x double> %v
+}
+
+; CHECK-LABEL: nearbyint_v2f64:
+; CHECK: f64.nearest
+declare <2 x double> @llvm.nearbyint.v2f64(<2 x double>)
+define <2 x double> @nearbyint_v2f64(<2 x double> %x) {
+  %v = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %x)
+  ret <2 x double> %v
+}
+
+; CHECK-LABEL: copysign_v2f64:
+; CHECK: f64.copysign
+declare <2 x double> @llvm.copysign.v2f64(<2 x double>, <2 x double>)
+define <2 x double> @copysign_v2f64(<2 x double> %x, <2 x double> %y) {
+  %v = call <2 x double> @llvm.copysign.v2f64(<2 x double> %x, <2 x double> %y)
+  ret <2 x double> %v
+}