Generate sexp instruction parser (#1754)

Also fix broken tests surfaced by the new parser.

Author: tlively

.flake8

@@ -1,3 +1,8 @@
 [flake8]
-ignore = E111,E114,E501,E121
+ignore =
+  E111, # indentation not a multiple of 4 (we use 2))
+  E114, # comment indentation not a multiple of 4
+  E501, # line too long
+  E121, # continuation line under-indented for hanging indent
+  E241  # space after comma (ignored for list in gen-s-parser.py)
 exclude = ./test/emscripten,./test/spec,./test/wasm-install

.travis.yml

@@ -15,7 +15,9 @@ jobs:
     - &test-ubuntu
       stage: test
       compiler: clang
-      python: 2.7
+      python:
+        - 2.7
+        - 3.6
       addons:
         apt:
           sources: ['ubuntu-toolchain-r-test']
@@ -33,6 +35,8 @@ jobs:
       script:
         - set -o errexit
         - flake8
+        # ensure generated parser is up to date
+        - ./scripts/gen-s-parser.py | diff src/gen-s-parser.inc -
         - BUILD_SUBDIR=${BUILD_SUBDIR:-.}
         - mkdir -p ${BUILD_SUBDIR} && cd ${BUILD_SUBDIR}
         - cmake ${TRAVIS_BUILD_DIR} -DCMAKE_C_FLAGS="$COMPILER_FLAGS" -DCMAKE_CXX_FLAGS="$COMPILER_FLAGS" -DCMAKE_INSTALL_PREFIX=install

scripts/gen-s-parser.py

@@ -167,8 +167,10 @@ class SExpressionWasmBuilder {
 
 private:
   Expression* makeExpression(Element& s);
-  Expression* makeBinary(Element& s, BinaryOp op, Type type);
-  Expression* makeUnary(Element& s, UnaryOp op, Type type);
+  Expression* makeUnreachable();
+  Expression* makeNop();
+  Expression* makeBinary(Element& s, BinaryOp op);
+  Expression* makeUnary(Element& s, UnaryOp op);
   Expression* makeSelect(Element& s);
   Expression* makeDrop(Element& s);
   Expression* makeHost(Element& s, HostOp op);

src/gen-s-parser.inc

@@ -658,241 +658,19 @@ Expression* SExpressionWasmBuilder::parseExpression(Element& s) {
 }
 
 Expression* SExpressionWasmBuilder::makeExpression(Element& s) {
-  IString id = s[0]->str();
-  const char *str = id.str;
-  const char *dot = strchr(str, '.');
-  if (dot) {
-    // type.operation (e.g. i32.add)
-    Type type = stringToType(str, false, true);
-    // Local copy to index into op without bounds checking.
-    enum { maxNameSize = 15 };
-    char op[maxNameSize + 1] = {'\0'};
-    strncpy(op, dot + 1, maxNameSize);
-#define BINARY_INT_OR_FLOAT(op) (type == i32 ? BinaryOp::op##Int32 : (type == i64 ? BinaryOp::op##Int64 : (type == f32 ? BinaryOp::op##Float32 : BinaryOp::op##Float64)))
-#define BINARY_INT(op) (type == i32 ? BinaryOp::op##Int32 : BinaryOp::op##Int64)
-#define BINARY_FLOAT(op) (type == f32 ? BinaryOp::op##Float32 : BinaryOp::op##Float64)
-    switch (op[0]) {
-      case 'a': {
-        if (op[1] == 'b') return makeUnary(s, type == f32 ? UnaryOp::AbsFloat32 : UnaryOp::AbsFloat64, type);
-        if (op[1] == 'd') return makeBinary(s, BINARY_INT_OR_FLOAT(Add), type);
-        if (op[1] == 'n') return makeBinary(s, BINARY_INT(And), type);
-        if (op[1] == 't' && !strncmp(op, "atomic.", strlen("atomic."))) {
-          if (op[7] == 'l') return makeLoad(s, type, /*isAtomic=*/true);
-          if (op[7] == 's') return makeStore(s, type, /*isAtomic=*/true);
-          if (op[7] == 'r') return makeAtomicRMWOrCmpxchg(s, type);
-        }
-        abort_on(op);
-      }
-      case 'c': {
-        if (op[1] == 'e') return makeUnary(s, type == f32 ? UnaryOp::CeilFloat32 : UnaryOp::CeilFloat64, type);
-        if (op[1] == 'l') return makeUnary(s, type == i32 ? UnaryOp::ClzInt32 : UnaryOp::ClzInt64, type);
-        if (op[1] == 'o') {
-          if (op[2] == 'p') return makeBinary(s, BINARY_FLOAT(CopySign), type);
-          if (op[2] == 'n') {
-            if (op[3] == 'v') {
-              if (op[8] == 's') return makeUnary(s, op[11] == '3' ? (type == f32 ? UnaryOp::ConvertSInt32ToFloat32 : UnaryOp::ConvertSInt32ToFloat64) : (type == f32 ? UnaryOp::ConvertSInt64ToFloat32 : UnaryOp::ConvertSInt64ToFloat64), type);
-              if (op[8] == 'u') return makeUnary(s, op[11] == '3' ? (type == f32 ? UnaryOp::ConvertUInt32ToFloat32 : UnaryOp::ConvertUInt32ToFloat64) : (type == f32 ? UnaryOp::ConvertUInt64ToFloat32 : UnaryOp::ConvertUInt64ToFloat64), type);
-            }
-            if (op[3] == 's') return makeConst(s, type);
-          }
-        }
-        if (op[1] == 't') return makeUnary(s, type == i32 ? UnaryOp::CtzInt32 : UnaryOp::CtzInt64, type);
-        abort_on(op);
-      }
-      case 'd': {
-        if (op[1] == 'i') {
-          if (op[3] == '_') return makeBinary(s, op[4] == 'u' ? BINARY_INT(DivU) : BINARY_INT(DivS), type);
-          if (op[3] == 0) return makeBinary(s, BINARY_FLOAT(Div), type);
-        }
-        if (op[1] == 'e') return makeUnary(s, UnaryOp::DemoteFloat64, type);
-        abort_on(op);
-      }
-      case 'e': {
-        if (op[1] == 'q') {
-          if (op[2] == 0) return makeBinary(s, BINARY_INT_OR_FLOAT(Eq), type);
-          if (op[2] == 'z') return makeUnary(s, type == i32 ? UnaryOp::EqZInt32 : UnaryOp::EqZInt64, type);
-        }
-        if (op[1] == 'x') {
-          if (op[6] == '8') return makeUnary(s, type == i32 ? UnaryOp::ExtendS8Int32 : UnaryOp::ExtendS8Int64, type);
-          if (op[6] == '1') return makeUnary(s, type == i32 ? UnaryOp::ExtendS16Int32 : UnaryOp::ExtendS16Int64, type);
-          if (op[6] == '3') return makeUnary(s, UnaryOp::ExtendS32Int64, type);
-          return makeUnary(s, op[7] == 'u' ? UnaryOp::ExtendUInt32 : UnaryOp::ExtendSInt32, type);
-        }
-        abort_on(op);
-      }
-      case 'f': {
-        if (op[1] == 'l') return makeUnary(s, type == f32 ? UnaryOp::FloorFloat32 : UnaryOp::FloorFloat64, type);
-        abort_on(op);
-      }
-      case 'g': {
-        if (op[1] == 't') {
-          if (op[2] == '_') return makeBinary(s, op[3] == 'u' ? BINARY_INT(GtU) : BINARY_INT(GtS), type);
-          if (op[2] == 0) return makeBinary(s, BINARY_FLOAT(Gt), type);
-        }
-        if (op[1] == 'e') {
-          if (op[2] == '_') return makeBinary(s, op[3] == 'u' ? BINARY_INT(GeU) : BINARY_INT(GeS), type);
-          if (op[2] == 0) return makeBinary(s, BINARY_FLOAT(Ge), type);
-        }
-        abort_on(op);
-      }
-      case 'l': {
-        if (op[1] == 't') {
-          if (op[2] == '_') return makeBinary(s, op[3] == 'u' ? BINARY_INT(LtU) : BINARY_INT(LtS), type);
-          if (op[2] == 0) return makeBinary(s, BINARY_FLOAT(Lt), type);
-        }
-        if (op[1] == 'e') {
-          if (op[2] == '_') return makeBinary(s, op[3] == 'u' ? BINARY_INT(LeU) : BINARY_INT(LeS), type);
-          if (op[2] == 0) return makeBinary(s, BINARY_FLOAT(Le), type);
-        }
-        if (op[1] == 'o') return makeLoad(s, type, /*isAtomic=*/false);
-        abort_on(op);
-      }
-      case 'm': {
-        if (op[1] == 'i') return makeBinary(s, BINARY_FLOAT(Min), type);
-        if (op[1] == 'a') return makeBinary(s, BINARY_FLOAT(Max), type);
-        if (op[1] == 'u') return makeBinary(s, BINARY_INT_OR_FLOAT(Mul), type);
-        abort_on(op);
-      }
-      case 'n': {
-        if (op[1] == 'e') {
-          if (op[2] == 0) return makeBinary(s, BINARY_INT_OR_FLOAT(Ne), type);
-          if (op[2] == 'a') return makeUnary(s, type == f32 ? UnaryOp::NearestFloat32 : UnaryOp::NearestFloat64, type);
-          if (op[2] == 'g') return makeUnary(s, type == f32 ? UnaryOp::NegFloat32 : UnaryOp::NegFloat64, type);
-        }
-        abort_on(op);
-      }
-      case 'o': {
-        if (op[1] == 'r') return makeBinary(s, BINARY_INT(Or), type);
-        abort_on(op);
-      }
-      case 'p': {
-        if (op[1] == 'r') return makeUnary(s,  UnaryOp::PromoteFloat32, type);
-        if (op[1] == 'o') return makeUnary(s, type == i32 ? UnaryOp::PopcntInt32 : UnaryOp::PopcntInt64, type);
-        abort_on(op);
-      }
-      case 'r': {
-        if (op[1] == 'e') {
-          if (op[2] == 'm') return makeBinary(s, op[4] == 'u' ? BINARY_INT(RemU) : BINARY_INT(RemS), type);
-          if (op[2] == 'i') return makeUnary(s, isFloatType(type) ? (type == f32 ? UnaryOp::ReinterpretInt32 : UnaryOp::ReinterpretInt64) : (type == i32 ? UnaryOp::ReinterpretFloat32 : UnaryOp::ReinterpretFloat64), type);
-        }
-        if (op[1] == 'o' && op[2] == 't') {
-          return makeBinary(s, op[3] == 'l' ? BINARY_INT(RotL) : BINARY_INT(RotR), type);
-        }
-        abort_on(op);
-      }
-      case 's': {
-        if (op[1] == 'h') {
-          if (op[2] == 'l') return makeBinary(s, BINARY_INT(Shl), type);
-          return makeBinary(s, op[4] == 'u' ? BINARY_INT(ShrU) : BINARY_INT(ShrS), type);
-        }
-        if (op[1] == 'u') return makeBinary(s, BINARY_INT_OR_FLOAT(Sub), type);
-        if (op[1] == 'q') return makeUnary(s, type == f32 ? UnaryOp::SqrtFloat32 : UnaryOp::SqrtFloat64, type);
-        if (op[1] == 't') return makeStore(s, type, /*isAtomic=*/false);
-        abort_on(op);
-      }
-      case 't': {
-        if (op[1] == 'r') {
-          if (op[6] == 's') return makeUnary(s, op[9] == '3' ? (type == i32 ? UnaryOp::TruncSFloat32ToInt32 : UnaryOp::TruncSFloat32ToInt64) : (type == i32 ? UnaryOp::TruncSFloat64ToInt32 : UnaryOp::TruncSFloat64ToInt64), type);
-          if (op[6] == 'u') return makeUnary(s, op[9] == '3' ? (type == i32 ? UnaryOp::TruncUFloat32ToInt32 : UnaryOp::TruncUFloat32ToInt64) : (type == i32 ? UnaryOp::TruncUFloat64ToInt32 : UnaryOp::TruncUFloat64ToInt64), type);
-          if (op[2] == 'u') return makeUnary(s, type == f32 ? UnaryOp::TruncFloat32 : UnaryOp::TruncFloat64, type);
-        }
-        abort_on(op);
-      }
-      case 'w': {
-        if (!strncmp(op, "wait", strlen("wait"))) return makeAtomicWait(s, type);
-        if (op[1] == 'r') return makeUnary(s,  UnaryOp::WrapInt64, type);
-        abort_on(op);
-      }
-      case 'x': {
-        if (op[1] == 'o') return makeBinary(s, BINARY_INT(Xor), type);
-        abort_on(op);
-      }
-      default: abort_on(op);
-    }
-  } else {
-    // other expression
-    switch (str[0]) {
-      case 'b': {
-        if (str[1] == 'l') return makeBlock(s);
-        if (str[1] == 'r') {
-          if (str[2] == '_' && str[3] == 't') return makeBreakTable(s);
-          return makeBreak(s);
-        }
-        abort_on(str);
-      }
-      case 'c': {
-        if (str[1] == 'a') {
-          if (id == CALL) return makeCall(s);
-          if (id == CALL_INDIRECT) return makeCallIndirect(s);
-        } else if (str[1] == 'u') return makeHost(s, HostOp::CurrentMemory);
-        abort_on(str);
-      }
-      case 'd': {
-        if (str[1] == 'r') return makeDrop(s);
-        abort_on(str);
-      }
-      case 'e': {
-        if (str[1] == 'l') return makeThenOrElse(s);
-        abort_on(str);
-      }
-      case 'g': {
-        if (str[1] == 'e') {
-          if (str[4] == 'l') return makeGetLocal(s);
-          if (str[4] == 'g') return makeGetGlobal(s);
-        }
-        if (str[1] == 'r') return makeHost(s, HostOp::GrowMemory);
-        abort_on(str);
-      }
-      case 'h': {
-        abort_on(str);
-      }
-      case 'i': {
-        if (str[1] == 'f') return makeIf(s);
-        abort_on(str);
-      }
-      case 'l': {
-        if (str[1] == 'o') return makeLoop(s);
-        abort_on(str);
-      }
-      case 'n': {
-        if (str[1] == 'o') return allocator.alloc<Nop>();
-        abort_on(str);
-      }
-      case 'p': {
-        abort_on(str);
-      }
-      case 's': {
-        if (str[1] == 'e' && str[2] == 't') {
-          if (str[4] == 'l') return makeSetLocal(s);
-          if (str[4] == 'g') return makeSetGlobal(s);
-        }
-        if (str[1] == 'e' && str[2] == 'l') return makeSelect(s);
-        abort_on(str);
-      }
-      case 'r': {
-        if (str[1] == 'e') return makeReturn(s);
-        abort_on(str);
-      }
-      case 't': {
-        if (str[1] == 'h') return makeThenOrElse(s);
-        if (str[1] == 'e' && str[2] == 'e') return makeTeeLocal(s);
-        abort_on(str);
-      }
-      case 'u': {
-        if (str[1] == 'n') return allocator.alloc<Unreachable>();
-        abort_on(str);
-      }
-      case 'w': {
-        if (!strncmp(str, "wake", strlen("wake"))) return makeAtomicWake(s);
-        abort_on(str);
-      }
-      default: abort_on(str);
-    }
-  }
-  abort_on("unrecognized input string for parsing");
+#define INSTRUCTION_PARSER
+#include "gen-s-parser.inc"
 }
 
-Expression* SExpressionWasmBuilder::makeBinary(Element& s, BinaryOp op, Type type) {
+Expression* SExpressionWasmBuilder::makeUnreachable() {
+  return allocator.alloc<Unreachable>();
+}
+
+Expression* SExpressionWasmBuilder::makeNop() {
+  return allocator.alloc<Nop>();
+}
+
+Expression* SExpressionWasmBuilder::makeBinary(Element& s, BinaryOp op) {
   auto ret = allocator.alloc<Binary>();
   ret->op = op;
   ret->left = parseExpression(s[1]);
@@ -902,67 +680,11 @@ Expression* SExpressionWasmBuilder::makeBinary(Element& s, BinaryOp op, Type typ
 }
 
 
-Expression* SExpressionWasmBuilder::makeUnary(Element& s, UnaryOp op, Type type) {
+Expression* SExpressionWasmBuilder::makeUnary(Element& s, UnaryOp op) {
   auto ret = allocator.alloc<Unary>();
   ret->op = op;
   ret->value = parseExpression(s[1]);
   ret->finalize();
-  // type is the reported type, e.g. i64.ctz reports i64 (but has a return type of i32, in this case)
-  // verify the reported type is correct
-  switch (op) {
-    case EqZInt32:
-    case NegFloat32:
-    case AbsFloat32:
-    case CeilFloat32:
-    case FloorFloat32:
-    case TruncFloat32:
-    case NearestFloat32:
-    case SqrtFloat32:
-    case ClzInt32:
-    case CtzInt32:
-    case PopcntInt32:
-    case EqZInt64:
-    case NegFloat64:
-    case AbsFloat64:
-    case CeilFloat64:
-    case FloorFloat64:
-    case TruncFloat64:
-    case NearestFloat64:
-    case SqrtFloat64:
-    case ClzInt64:
-    case CtzInt64:
-    case PopcntInt64: {
-      if (ret->value->type != unreachable && type != ret->value->type) throw ParseException(std::string("bad type for ") + getExpressionName(ret) + ": " + printType(type) + " vs value type " + printType(ret->value->type), s.line, s.col);
-      break;
-    }
-    case ExtendSInt32: case ExtendUInt32:
-    case ExtendS8Int32: case ExtendS16Int32:
-    case ExtendS8Int64: case ExtendS16Int64: case ExtendS32Int64:
-    case WrapInt64:
-    case PromoteFloat32:
-    case DemoteFloat64:
-    case TruncSFloat32ToInt32:
-    case TruncUFloat32ToInt32:
-    case TruncSFloat64ToInt32:
-    case TruncUFloat64ToInt32:
-    case ReinterpretFloat32:
-    case TruncSFloat32ToInt64:
-    case TruncUFloat32ToInt64:
-    case TruncSFloat64ToInt64:
-    case TruncUFloat64ToInt64:
-    case ReinterpretFloat64:
-    case ReinterpretInt32:
-    case ConvertSInt32ToFloat32:
-    case ConvertUInt32ToFloat32:
-    case ConvertSInt64ToFloat32:
-    case ConvertUInt64ToFloat32:
-    case ReinterpretInt64:
-    case ConvertSInt32ToFloat64:
-    case ConvertUInt32ToFloat64:
-    case ConvertSInt64ToFloat64:
-    case ConvertUInt64ToFloat64: break;
-    default: WASM_UNREACHABLE();
-  }
   return ret;
 }
 

src/wasm-s-parser.h

@@ -17,8 +17,10 @@
   )
   (func $test2
     (drop (i32.load (i32.const 12))) ;; a safe load
-    (drop (i32.load16 (i32.const 12)))
-    (drop (i32.load8 (i32.const 12)))
+    (drop (i32.load16_s (i32.const 12)))
+    (drop (i32.load8_s (i32.const 12)))
+    (drop (i32.load16_u (i32.const 12)))
+    (drop (i32.load8_u (i32.const 12)))
   )
   (func $test3
     (i32.store (i32.const 12) (i32.const 115)) ;; a safe store, should alter memory

src/wasm/wasm-s-parser.cpp

@@ -14,8 +14,10 @@
   )
   (func $test2
     (drop (i32.load (i32.const 12))) ;; a safe load
-    (drop (i32.load16 (i32.const 12)))
-    (drop (i32.load8 (i32.const 12)))
+    (drop (i32.load16_s (i32.const 12)))
+    (drop (i32.load8_s (i32.const 12)))
+    (drop (i32.load16_u (i32.const 12)))
+    (drop (i32.load8_u (i32.const 12)))
   )
   (func $test3
     (i32.store (i32.const 12) (i32.const 115)) ;; a safe store, should alter memory

test/ctor-eval/basics-flatten.wast

@@ -733,7 +733,7 @@
   (memory 10)
   (type $0 (func))
   (func $keep2 (type $0)
-    (i32.store32 offset=3
+    (i32.store offset=3
       (i32.const 0)
       (i32.const 100)
     )