From be6f50798e79336cdfd8fe464f37d41ac135640d Mon Sep 17 00:00:00 2001
From: Thomas Lively <tlively@google.com>
Date: Thu, 29 Oct 2020 11:06:20 -0700
Subject: [PATCH] [WebAssembly] Implement SIMD signselect instructions

As proposed in https://github.com/WebAssembly/simd/pull/124, using the opcodes
adopted by V8 in
https://chromium-review.googlesource.com/c/v8/v8/+/2486235/2/src/wasm/wasm-opcodes.h.
Uses new builtin functions and a new target intrinsic exclusively to ensure that
the new instructions are only emitted when a user explicitly opts in to using
them since they are still in the prototyping and evaluation phase.

Differential Revision: https://reviews.llvm.org/D90357
---
 .../clang/Basic/BuiltinsWebAssembly.def       |  6 +++
 clang/lib/CodeGen/CGBuiltin.cpp               | 11 +++++
 clang/test/CodeGen/builtins-wasm.c            | 28 +++++++++++
 llvm/include/llvm/IR/IntrinsicsWebAssembly.td |  5 ++
 .../WebAssembly/WebAssemblyInstrSIMD.td       | 17 +++++++
 .../CodeGen/WebAssembly/simd-intrinsics.ll    | 48 +++++++++++++++++++
 llvm/test/MC/WebAssembly/simd-encodings.s     | 12 +++++
 7 files changed, 127 insertions(+)

diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index e86346d71c59..c1e594e147b3 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -134,6 +134,12 @@ TARGET_BUILTIN(__builtin_wasm_extmul_low_i32x4_u_i64x2, "V2ULLiV4UiV4Ui", "nc",
 TARGET_BUILTIN(__builtin_wasm_extmul_high_i32x4_u_i64x2, "V2ULLiV4UiV4Ui", "nc", "simd128")
 
 TARGET_BUILTIN(__builtin_wasm_bitselect, "V4iV4iV4iV4i", "nc", "simd128")
+
+TARGET_BUILTIN(__builtin_wasm_signselect_i8x16, "V16ScV16ScV16ScV16Sc", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_signselect_i16x8, "V8sV8sV8sV8s", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_signselect_i32x4, "V4iV4iV4iV4i", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_signselect_i64x2, "V2LLiV2LLiV2LLiV2LLi", "nc", "simd128")
+
 TARGET_BUILTIN(__builtin_wasm_shuffle_v8x16, "V16ScV16ScV16ScIiIiIiIiIiIiIiIiIiIiIiIiIiIiIiIi", "nc", "simd128")
 
 TARGET_BUILTIN(__builtin_wasm_any_true_i8x16, "iV16Sc", "nc", "simd128")
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index f933113fa883..7341a440b873 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -16675,6 +16675,17 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
         CGM.getIntrinsic(Intrinsic::wasm_bitselect, ConvertType(E->getType()));
     return Builder.CreateCall(Callee, {V1, V2, C});
   }
+  case WebAssembly::BI__builtin_wasm_signselect_i8x16:
+  case WebAssembly::BI__builtin_wasm_signselect_i16x8:
+  case WebAssembly::BI__builtin_wasm_signselect_i32x4:
+  case WebAssembly::BI__builtin_wasm_signselect_i64x2: {
+    Value *V1 = EmitScalarExpr(E->getArg(0));
+    Value *V2 = EmitScalarExpr(E->getArg(1));
+    Value *C = EmitScalarExpr(E->getArg(2));
+    Function *Callee =
+        CGM.getIntrinsic(Intrinsic::wasm_signselect, ConvertType(E->getType()));
+    return Builder.CreateCall(Callee, {V1, V2, C});
+  }
   case WebAssembly::BI__builtin_wasm_dot_s_i32x4_i16x8: {
     Value *LHS = EmitScalarExpr(E->getArg(0));
     Value *RHS = EmitScalarExpr(E->getArg(1));
diff --git a/clang/test/CodeGen/builtins-wasm.c b/clang/test/CodeGen/builtins-wasm.c
index ee635317ff7a..e5c7211ad5be 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -622,6 +622,34 @@ i32x4 bitselect(i32x4 x, i32x4 y, i32x4 c) {
   // WEBASSEMBLY-NEXT: ret
 }
 
+i8x16 signselect_i8x16(i8x16 x, i8x16 y, i8x16 c) {
+  return __builtin_wasm_signselect_i8x16(x, y, c);
+  // WEBASSEMBLY: call <16 x i8> @llvm.wasm.signselect.v16i8(
+  // WEBASSEMBLY-SAME: <16 x i8> %x, <16 x i8> %y, <16 x i8> %c)
+  // WEBASSEMBLY-NEXT: ret
+}
+
+i16x8 signselect_i16x8(i16x8 x, i16x8 y, i16x8 c) {
+  return __builtin_wasm_signselect_i16x8(x, y, c);
+  // WEBASSEMBLY: call <8 x i16> @llvm.wasm.signselect.v8i16(
+  // WEBASSEMBLY-SAME: <8 x i16> %x, <8 x i16> %y, <8 x i16> %c)
+  // WEBASSEMBLY-NEXT: ret
+}
+
+i32x4 signselect_i32x4(i32x4 x, i32x4 y, i32x4 c) {
+  return __builtin_wasm_signselect_i32x4(x, y, c);
+  // WEBASSEMBLY: call <4 x i32> @llvm.wasm.signselect.v4i32(
+  // WEBASSEMBLY-SAME: <4 x i32> %x, <4 x i32> %y, <4 x i32> %c)
+  // WEBASSEMBLY-NEXT: ret
+}
+
+i64x2 signselect_i64x2(i64x2 x, i64x2 y, i64x2 c) {
+  return __builtin_wasm_signselect_i64x2(x, y, c);
+  // WEBASSEMBLY: call <2 x i64> @llvm.wasm.signselect.v2i64(
+  // WEBASSEMBLY-SAME: <2 x i64> %x, <2 x i64> %y, <2 x i64> %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)
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index 5c503a4d6436..f65b7457436b 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -276,6 +276,11 @@ def int_wasm_extmul_high_unsigned :
             [LLVMSubdivide2VectorType<0>, LLVMSubdivide2VectorType<0>],
             [IntrNoMem, IntrSpeculatable]>;
 
+def int_wasm_signselect :
+  Intrinsic<[llvm_anyvector_ty],
+            [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
+            [IntrNoMem, IntrSpeculatable]>;
+
 //===----------------------------------------------------------------------===//
 // Thread-local storage intrinsics
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
index 7e78b6c0eecd..2d8c8160641e 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
@@ -747,6 +747,23 @@ def : Pat<(select
           )>;
 } // foreach vec_t
 
+// Sign select
+multiclass SIMDSignSelect<ValueType vec_t, string vec, bits<32> simdop> {
+  defm SIGNSELECT_#vec_t :
+    SIMD_I<(outs V128:$dst), (ins V128:$v1, V128:$v2, V128:$c), (outs), (ins),
+           [(set (vec_t V128:$dst),
+             (vec_t (int_wasm_signselect
+               (vec_t V128:$v1), (vec_t V128:$v2), (vec_t V128:$c)
+             ))
+           )],
+           vec#".signselect\t$dst, $v1, $v2, $c", vec#".signselect", simdop>;
+}
+
+defm : SIMDSignSelect<v16i8, "i8x16", 125>;
+defm : SIMDSignSelect<v8i16, "i16x8", 126>;
+defm : SIMDSignSelect<v4i32, "i32x4", 127>;
+defm : SIMDSignSelect<v2i64, "i64x2", 148>;
+
 //===----------------------------------------------------------------------===//
 // Integer unary arithmetic
 //===----------------------------------------------------------------------===//
diff --git a/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll b/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll
index b4edbd3daee2..aa4ff63b9676 100644
--- a/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll
+++ b/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll
@@ -127,6 +127,18 @@ define <16 x i8> @bitselect_v16i8(<16 x i8> %v1, <16 x i8> %v2, <16 x i8> %c) {
   ret <16 x i8> %a
 }
 
+; CHECK-LABEL: signselect_v16i8:
+; SIMD128-NEXT: .functype signselect_v16i8 (v128, v128, v128) -> (v128){{$}}
+; SIMD128-NEXT: i8x16.signselect $push[[R:[0-9]+]]=, $0, $1, $2{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <16 x i8> @llvm.wasm.signselect.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)
+define <16 x i8> @signselect_v16i8(<16 x i8> %v1, <16 x i8> %v2, <16 x i8> %c) {
+  %a = call <16 x i8> @llvm.wasm.signselect.v16i8(
+     <16 x i8> %v1, <16 x i8> %v2, <16 x i8> %c
+  )
+  ret <16 x i8> %a
+}
+
 ; CHECK-LABEL: narrow_signed_v16i8:
 ; SIMD128-NEXT: .functype narrow_signed_v16i8 (v128, v128) -> (v128){{$}}
 ; SIMD128-NEXT: i8x16.narrow_i16x8_s $push[[R:[0-9]+]]=, $0, $1{{$}}
@@ -339,6 +351,18 @@ define <8 x i16> @bitselect_v8i16(<8 x i16> %v1, <8 x i16> %v2, <8 x i16> %c) {
   ret <8 x i16> %a
 }
 
+; CHECK-LABEL: signselect_v8i16:
+; SIMD128-NEXT: .functype signselect_v8i16 (v128, v128, v128) -> (v128){{$}}
+; SIMD128-NEXT: i16x8.signselect $push[[R:[0-9]+]]=, $0, $1, $2{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <8 x i16> @llvm.wasm.signselect.v8i16(<8 x i16>, <8 x i16>, <8 x i16>)
+define <8 x i16> @signselect_v8i16(<8 x i16> %v1, <8 x i16> %v2, <8 x i16> %c) {
+  %a = call <8 x i16> @llvm.wasm.signselect.v8i16(
+    <8 x i16> %v1, <8 x i16> %v2, <8 x i16> %c
+  )
+  ret <8 x i16> %a
+}
+
 ; CHECK-LABEL: narrow_signed_v8i16:
 ; SIMD128-NEXT: .functype narrow_signed_v8i16 (v128, v128) -> (v128){{$}}
 ; SIMD128-NEXT: i16x8.narrow_i32x4_s $push[[R:[0-9]+]]=, $0, $1{{$}}
@@ -467,6 +491,18 @@ define <4 x i32> @bitselect_v4i32(<4 x i32> %v1, <4 x i32> %v2, <4 x i32> %c) {
   ret <4 x i32> %a
 }
 
+; CHECK-LABEL: signselect_v4i32:
+; SIMD128-NEXT: .functype signselect_v4i32 (v128, v128, v128) -> (v128){{$}}
+; SIMD128-NEXT: i32x4.signselect $push[[R:[0-9]+]]=, $0, $1, $2{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <4 x i32> @llvm.wasm.signselect.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+define <4 x i32> @signselect_v4i32(<4 x i32> %v1, <4 x i32> %v2, <4 x i32> %c) {
+  %a = call <4 x i32> @llvm.wasm.signselect.v4i32(
+    <4 x i32> %v1, <4 x i32> %v2, <4 x i32> %c
+  )
+  ret <4 x i32> %a
+}
+
 ; CHECK-LABEL: trunc_sat_s_v4i32:
 ; NO-SIMD128-NOT: f32x4
 ; SIMD128-NEXT: .functype trunc_sat_s_v4i32 (v128) -> (v128){{$}}
@@ -572,6 +608,18 @@ define <2 x i64> @bitselect_v2i64(<2 x i64> %v1, <2 x i64> %v2, <2 x i64> %c) {
   ret <2 x i64> %a
 }
 
+; CHECK-LABEL: signselect_v2i64:
+; SIMD128-NEXT: .functype signselect_v2i64 (v128, v128, v128) -> (v128){{$}}
+; SIMD128-NEXT: i64x2.signselect $push[[R:[0-9]+]]=, $0, $1, $2{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <2 x i64> @llvm.wasm.signselect.v2i64(<2 x i64>, <2 x i64>, <2 x i64>)
+define <2 x i64> @signselect_v2i64(<2 x i64> %v1, <2 x i64> %v2, <2 x i64> %c) {
+  %a = call <2 x i64> @llvm.wasm.signselect.v2i64(
+    <2 x i64> %v1, <2 x i64> %v2, <2 x i64> %c
+  )
+  ret <2 x i64> %a
+}
+
 ; ==============================================================================
 ; 4 x f32
 ; ==============================================================================
diff --git a/llvm/test/MC/WebAssembly/simd-encodings.s b/llvm/test/MC/WebAssembly/simd-encodings.s
index a4908a0a61af..6bd54c9ddeed 100644
--- a/llvm/test/MC/WebAssembly/simd-encodings.s
+++ b/llvm/test/MC/WebAssembly/simd-encodings.s
@@ -694,4 +694,16 @@ main:
     # CHECK: i64x2.extmul_high_i32x4_u # encoding: [0xfd,0xd7,0x01]
     i64x2.extmul_high_i32x4_u
 
+    # CHECK: i8x16.signselect # encoding: [0xfd,0x7d]
+    i8x16.signselect
+
+    # CHECK: i16x8.signselect # encoding: [0xfd,0x7e]
+    i16x8.signselect
+
+    # CHECK: i32x4.signselect # encoding: [0xfd,0x7f]
+    i32x4.signselect
+
+    # CHECK: i64x2.signselect # encoding: [0xfd,0x94,0x01]
+    i64x2.signselect
+
     end_function