i64Math in asm

5 files changed, 33 insertions, 40 deletions

diff --git a/emscripten.py b/emscripten.py
index c0c40da7..7c064850 100755
--- a/emscripten.py
+++ b/emscripten.py
@@ -291,11 +291,19 @@ def emscript(infile, settings, outfile, libraries=[]):
 
   function_tables_defs = '\n'.join([table for table in last_forwarded_json['Functions']['tables'].itervalues()])
   if settings.get('ASM_JS'):
+    asm_setup = ''
     fundamentals = ['buffer', 'Int8Array', 'Int16Array', 'Int32Array', 'Uint8Array', 'Uint16Array', 'Uint32Array', 'Float32Array', 'Float64Array']
     basics = ['abort', 'assert', 'STACKTOP', 'STACK_MAX', 'tempDoublePtr', 'ABORT']
     if not settings['NAMED_GLOBALS']: basics += ['GLOBAL_BASE']
-    #if forwarded_json['Types']['preciseI64MathUsed']:
-    #  basics += ['i64Math']
+    if forwarded_json['Types']['preciseI64MathUsed']:
+      basics += ['i64Math_' + op for op in ['add', 'subtract', 'multiply', 'divide', 'modulo']]
+      asm_setup += '''
+var i64Math_add = function(a, b, c, d) { i64Math.add(a, b, c, d) };
+var i64Math_subtract = function(a, b, c, d) { i64Math.subtract(a, b, c, d) };
+var i64Math_multiply = function(a, b, c, d) { i64Math.multiply(a, b, c, d) };
+var i64Math_divide = function(a, b, c, d) { i64Math.divide(a, b, c, d) };
+var i64Math_modulo = function(a, b, c, d) { i64Math.modulo(a, b, c, d) };
+'''
     asm_runtime_funcs = ['stackAlloc', 'stackSave', 'stackRestore']
     # function tables
     function_tables = ['dynCall_' + table for table in last_forwarded_json['Functions']['tables']]
@@ -329,6 +337,7 @@ def emscript(infile, settings, outfile, libraries=[]):
     receiving = ';\n'.join(['var ' + s + ' = Module["' + s + '"] = asm.' + s for s in exported_implemented_functions + function_tables])
     # finalize
     funcs_js = '''
+%s
 var asmPre = (function(env, buffer) {
   'use asm';
   var HEAP8 = new env.Int8Array(buffer);
@@ -339,7 +348,7 @@ var asmPre = (function(env, buffer) {
   var HEAPU32 = new env.Uint32Array(buffer);
   var HEAPF32 = new env.Float32Array(buffer);
   var HEAPF64 = new env.Float64Array(buffer);
-''' + asm_globals + '''
+''' % (asm_setup,) + asm_globals + '''
   function stackAlloc(size) {
     var ret = STACKTOP;
     STACKTOP = (STACKTOP + size)|0;
diff --git a/src/library.js b/src/library.js
index f8e1022b..9e447232 100644
--- a/src/library.js
+++ b/src/library.js
@@ -5064,16 +5064,16 @@ LibraryManager.library = {
   llvm_uadd_with_overflow_i64: function(xl, xh, yl, yh) {
     i64Math.add(xl, xh, yl, yh);
     return {
-      f0: i64Math.result,
+      f0: [HEAP32[tempDoublePtr>>2], HEAP32[tempDoublePtr+4>>2]],
       f1: 0 // XXX Need to hack support for this in long.js
     };
   },
 
   llvm_umul_with_overflow_i64__deps: [function() { Types.preciseI64MathUsed = 1 }],
   llvm_umul_with_overflow_i64: function(xl, xh, yl, yh) {
-    i64Math.mul(xl, xh, yl, yh);
+    i64Math.multiply(xl, xh, yl, yh);
     return {
-      f0: i64Math.result,
+      f0: [HEAP32[tempDoublePtr>>2], HEAP32[tempDoublePtr+4>>2]],
       f1: 0 // XXX Need to hack support for this in long.js
     };
   },
diff --git a/src/long.js b/src/long.js
index d5770e48..865540db 100644
--- a/src/long.js
+++ b/src/long.js
@@ -1530,27 +1530,26 @@ var i64Math = (function() { // Emscripten wrapper
 
   // Emscripten wrapper
   var Wrapper = {
-    result: [0, 0], // return result stored here
     add: function(xl, xh, yl, yh) {
       var x = new goog.math.Long(xl, xh);
       var y = new goog.math.Long(yl, yh);
       var ret = x.add(y);
-      Wrapper.result[0] = ret.low_;
-      Wrapper.result[1] = ret.high_;
+      HEAP32[tempDoublePtr>>2] = ret.low_;
+      HEAP32[tempDoublePtr+4>>2] = ret.high_;
     },
     subtract: function(xl, xh, yl, yh) {
       var x = new goog.math.Long(xl, xh);
       var y = new goog.math.Long(yl, yh);
       var ret = x.subtract(y);
-      Wrapper.result[0] = ret.low_;
-      Wrapper.result[1] = ret.high_;
+      HEAP32[tempDoublePtr>>2] = ret.low_;
+      HEAP32[tempDoublePtr+4>>2] = ret.high_;
     },
     multiply: function(xl, xh, yl, yh) {
       var x = new goog.math.Long(xl, xh);
       var y = new goog.math.Long(yl, yh);
       var ret = x.multiply(y);
-      Wrapper.result[0] = ret.low_;
-      Wrapper.result[1] = ret.high_;
+      HEAP32[tempDoublePtr>>2] = ret.low_;
+      HEAP32[tempDoublePtr+4>>2] = ret.high_;
     },
     makeTwo32: function() {
       Wrapper.two32 = new BigInteger();
@@ -1573,8 +1572,8 @@ var i64Math = (function() { // Emscripten wrapper
         var x = new goog.math.Long(xl, xh);
         var y = new goog.math.Long(yl, yh);
         var ret = x.div(y);
-        Wrapper.result[0] = ret.low_;
-        Wrapper.result[1] = ret.high_;
+        HEAP32[tempDoublePtr>>2] = ret.low_;
+        HEAP32[tempDoublePtr+4>>2] = ret.high_;
       } else {
         // slow precise bignum division
         var x = Wrapper.lh2bignum(xl >>> 0, xh >>> 0);
@@ -1584,8 +1583,8 @@ var i64Math = (function() { // Emscripten wrapper
         var l = new BigInteger();
         var h = new BigInteger();
         z.divRemTo(Wrapper.two32, h, l);
-        Wrapper.result[0] = parseInt(l.toString()) | 0;
-        Wrapper.result[1] = parseInt(h.toString()) | 0;
+        HEAP32[tempDoublePtr>>2] = parseInt(l.toString()) | 0;
+        HEAP32[tempDoublePtr+4>>2] = parseInt(h.toString()) | 0;
       }
     },
     modulo: function(xl, xh, yl, yh, unsigned) {
@@ -1594,8 +1593,8 @@ var i64Math = (function() { // Emscripten wrapper
         var x = new goog.math.Long(xl, xh);
         var y = new goog.math.Long(yl, yh);
         var ret = x.modulo(y);
-        Wrapper.result[0] = ret.low_;
-        Wrapper.result[1] = ret.high_;
+        HEAP32[tempDoublePtr>>2] = ret.low_;
+        HEAP32[tempDoublePtr+4>>2] = ret.high_;
       } else {
         // slow precise bignum division
         var x = Wrapper.lh2bignum(xl >>> 0, xh >>> 0);
@@ -1605,8 +1604,8 @@ var i64Math = (function() { // Emscripten wrapper
         var l = new BigInteger();
         var h = new BigInteger();
         z.divRemTo(Wrapper.two32, h, l);
-        Wrapper.result[0] = parseInt(l.toString()) | 0;
-        Wrapper.result[1] = parseInt(h.toString()) | 0;
+        HEAP32[tempDoublePtr>>2] = parseInt(l.toString()) | 0;
+        HEAP32[tempDoublePtr+4>>2] = parseInt(h.toString()) | 0;
       }
     },
     stringify: function(l, h, unsigned) {
diff --git a/src/parseTools.js b/src/parseTools.js
index e31a3027..258ef40a 100644
--- a/src/parseTools.js
+++ b/src/parseTools.js
@@ -1737,8 +1737,8 @@ function processMathop(item) {
     }
     function i64PreciseOp(type, lastArg) {
       Types.preciseI64MathUsed = true;
-      return finish(['(i64Math.' + type + '(' + low1 + ',' + high1 + ',' + low2 + ',' + high2 +
-                     (lastArg ? ',' + lastArg : '') + '),i64Math.result[0])', 'i64Math.result[1]']);
+      return finish(['(i64Math' + (ASM_JS ? '_' : '.') + type + '(' + low1 + ',' + high1 + ',' + low2 + ',' + high2 +
+                     (lastArg ? ',' + lastArg : '') + '),' + makeGetValue('tempDoublePtr', 0, 'i32') + ')', makeGetValue('tempDoublePtr', Runtime.getNativeTypeSize('i32'), 'i32')]);
     }
     switch (op) {
       // basic integer ops
@@ -1898,7 +1898,7 @@ function processMathop(item) {
     case 'mul': {
       if (bits == 32 && PRECISE_I32_MUL) {
         Types.preciseI64MathUsed = true;
-        return '(i64Math.multiply(' + idents[0] + ',0,' + idents[1] + ',0),i64Math.result[0])';
+        return '(i64Math' + (ASM_JS ? '_' : '.') + 'multiply(' + idents[0] + ',0,' + idents[1] + ',0),' + makeGetValue('tempDoublePtr', 0, 'i32') + ')';
       } else {
         return handleOverflow(getFastValue(idents[0], '*', idents[1], item.type), bits);
       }
diff --git a/tests/runner.py b/tests/runner.py
index 6c176254..fa7be147 100755
--- a/tests/runner.py
+++ b/tests/runner.py
@@ -487,8 +487,6 @@ if 'benchmark' not in str(sys.argv) and 'sanity' not in str(sys.argv) and 'brows
         self.do_run(src, 'hello, world!')
 
     def test_intvars(self):
-        if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
-
         src = '''
           #include <stdio.h>
           int global = 20;
@@ -574,7 +572,6 @@ if 'benchmark' not in str(sys.argv) and 'sanity' not in str(sys.argv) and 'brows
 
     def test_i64(self):
         if Settings.USE_TYPED_ARRAYS != 2: return self.skip('i64 mode 1 requires ta2')
-        if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
         src = '''
           #include <stdio.h>
@@ -790,7 +787,6 @@ if 'benchmark' not in str(sys.argv) and 'sanity' not in str(sys.argv) and 'brows
 
     def test_i64_b(self):
         if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-        if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
         src = r'''
           #include <stdio.h>
@@ -816,7 +812,6 @@ if 'benchmark' not in str(sys.argv) and 'sanity' not in str(sys.argv) and 'brows
 
     def test_i64_cmp(self):
         if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-        if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
         src = r'''
           #include <stdio.h>
@@ -841,7 +836,6 @@ if 'benchmark' not in str(sys.argv) and 'sanity' not in str(sys.argv) and 'brows
 
     def test_i64_cmp2(self):
         if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-        if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
         src = r'''
           #include <inttypes.h>
@@ -888,7 +882,7 @@ m_divisor is 1091269979
 
     def test_i64_double(self):
         if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-        if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
+
 
         src = r'''
           #include <stdio.h>
@@ -932,7 +926,6 @@ m_divisor is 1091269979
 
     def test_i64_umul(self):
         if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-        if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
         src = r'''
           #include <inttypes.h>
@@ -958,7 +951,6 @@ m_divisor is 1091269979
 
     def test_i64_precise(self):
         if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-        if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
         src = r'''
           #include <inttypes.h>
@@ -1039,7 +1031,6 @@ m_divisor is 1091269979
 
     def test_i64_zextneg(self):
       if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-      if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
       src = r'''
         #include <stdint.h>
@@ -1061,7 +1052,6 @@ m_divisor is 1091269979
 
     def test_i64_7z(self):
       if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-      if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
       src = r'''
         #include <stdint.h>
@@ -1083,7 +1073,6 @@ m_divisor is 1091269979
 
     def test_i64_i16(self):
       if Settings.USE_TYPED_ARRAYS != 2: return self.skip('full i64 stuff only in ta2')
-      if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
       src = r'''
         #include <stdio.h>
@@ -1126,7 +1115,6 @@ m_divisor is 1091269979
 
     def test_i32_mul_precise(self):
       if self.emcc_args == None: return self.skip('needs ta2')
-      if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
       self.emcc_args += ['-s', 'PRECISE_I32_MUL=1']
       src = r'''
@@ -1209,7 +1197,6 @@ c5,de,15,8a
       self.do_run(open(path_from_root('tests', 'cube2md5.cpp')).read(), open(path_from_root('tests', 'cube2md5.ok')).read())
 
     def test_cube2hash(self):
-      if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
 
       try:
         old_chunk_size = os.environ.get('EMSCRIPT_MAX_CHUNK_SIZE') or ''
@@ -7106,8 +7093,6 @@ def process(filename):
         assert 'Assertion failed' in str(e), str(e)
 
     def test_linespecific(self):
-      if Settings.ASM_JS: return self.skip('asm does not support i64 yet')
-
       if '-g' not in Building.COMPILER_TEST_OPTS: Building.COMPILER_TEST_OPTS.append('-g')
       if self.emcc_args: self.emcc_args += ['--llvm-opts', '0'] # llvm full opts make the expected failures here not happen