[WebAssembly] Prototype i8x16.popcnt

As proposed at https://github.com/WebAssembly/simd/pull/379. Use a target
builtin and intrinsic rather than normal codegen patterns to make the
instruction opt-in until it is merged to the proposal and stabilized in engines.

Differential Revision: https://reviews.llvm.org/D89446

clang/include/clang/Basic/BuiltinsWebAssembly.def

@@ -114,6 +114,8 @@ TARGET_BUILTIN(__builtin_wasm_max_u_i32x4, "V4UiV4UiV4Ui", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_avgr_u_i8x16, "V16UcV16UcV16Uc", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_avgr_u_i16x8, "V8UsV8UsV8Us", "nc", "simd128")
 
+TARGET_BUILTIN(__builtin_wasm_popcnt_i8x16, "V16ScV16Sc", "nc", "simd128")
+
 TARGET_BUILTIN(__builtin_wasm_q15mulr_saturate_s_i8x16, "V8sV8sV8s", "nc", "simd128")
 
 TARGET_BUILTIN(__builtin_wasm_bitselect, "V4iV4iV4iV4i", "nc", "simd128")

clang/lib/CodeGen/CGBuiltin.cpp

@@ -16606,6 +16606,11 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
     Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_dot);
     return Builder.CreateCall(Callee, {LHS, RHS});
   }
+  case WebAssembly::BI__builtin_wasm_popcnt_i8x16: {
+    Value *Vec = EmitScalarExpr(E->getArg(0));
+    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_popcnt);
+    return Builder.CreateCall(Callee, {Vec});
+  }
   case WebAssembly::BI__builtin_wasm_any_true_i8x16:
   case WebAssembly::BI__builtin_wasm_any_true_i16x8:
   case WebAssembly::BI__builtin_wasm_any_true_i32x4:

clang/test/CodeGen/builtins-wasm.c

@@ -538,6 +538,12 @@ i32x4 bitselect(i32x4 x, i32x4 y, i32x4 c) {
   // WEBASSEMBLY-NEXT: ret
 }
 
+i8x16 popcnt(i8x16 x) {
+  return __builtin_wasm_popcnt_i8x16(x);
+  // WEBASSEMBLY: call <16 x i8> @llvm.wasm.popcnt(<16 x i8> %x)
+  // WEBASSEMBLY-NEXT: ret
+}
+
 int any_true_i8x16(i8x16 x) {
   return __builtin_wasm_any_true_i8x16(x);
   // WEBASSEMBLY: call i32 @llvm.wasm.anytrue.v16i8(<16 x i8> %x)

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

@@ -254,6 +254,11 @@ def int_wasm_store64_lane :
             [IntrWriteMem, IntrArgMemOnly],
             "", [SDNPMemOperand]>;
 
+// TODO: Replace this intrinsic with normal ISel patterns once popcnt is merged
+// to the proposal.
+def int_wasm_popcnt :
+  Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty], [IntrNoMem, IntrSpeculatable]>;
+
 //===----------------------------------------------------------------------===//
 // Thread-local storage intrinsics
 //===----------------------------------------------------------------------===//

llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td

@@ -785,6 +785,9 @@ defm ANYTRUE : SIMDReduce<int_wasm_anytrue, "any_true", 98>;
 // All lanes true: all_true
 defm ALLTRUE : SIMDReduce<int_wasm_alltrue, "all_true", 99>;
 
+// Population count: popcnt
+defm POPCNT : SIMDUnary<v16i8, "i8x16", int_wasm_popcnt, "popcnt", 124>;
+
 // Reductions already return 0 or 1, so and 1, setne 0, and seteq 1
 // can be folded out
 foreach reduction =

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

@@ -75,6 +75,16 @@ define <16 x i8> @avgr_u_v16i8(<16 x i8> %x, <16 x i8> %y) {
   ret <16 x i8> %a
 }
 
+; CHECK-LABEL: popcnt_v16i8:
+; SIMD128-NEXT: .functype popcnt_v16i8 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i8x16.popcnt $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <16 x i8> @llvm.wasm.popcnt(<16 x i8>)
+define <16 x i8> @popcnt_v16i8(<16 x i8> %x) {
+ %a = call <16 x i8> @llvm.wasm.popcnt(<16 x i8> %x)
+ ret <16 x i8> %a
+}
+
 ; CHECK-LABEL: any_v16i8:
 ; SIMD128-NEXT: .functype any_v16i8 (v128) -> (i32){{$}}
 ; SIMD128-NEXT: i8x16.any_true $push[[R:[0-9]+]]=, $0{{$}}

llvm/test/MC/WebAssembly/simd-encodings.s

@@ -367,6 +367,9 @@ main:
     # CHECK: i8x16.avgr_u # encoding: [0xfd,0x7b]
     i8x16.avgr_u
 
+    # CHECK: i8x16.popcnt # encoding: [0xfd,0x7c]
+    i8x16.popcnt
+
     # CHECK: i16x8.abs # encoding: [0xfd,0x80,0x01]
     i16x8.abs