Merge branch 'master' into zwy2

Author: cjld

python/jittor/__init__.py

@@ -7,7 +7,7 @@
 # This file is subject to the terms and conditions defined in
 # file 'LICENSE.txt', which is part of this source code package.
 # ***************************************************************
-__version__ = '1.1.7.4'
+__version__ = '1.1.7.5'
 from . import lock
 with lock.lock_scope():
     from . import compiler

python/jittor/test/test_clone.py

@@ -25,7 +25,7 @@ def test2(self):
         a = jt.array([1,2])
         print(a.detach())
 
-    @jt.flag_scope(eager_execution=1)
+    @jt.flag_scope(lazy_execution=0)
     def test3(self):
         a = jt.array([1,2])
         print(a.detach())

python/jittor/test/test_function.py

@@ -284,10 +284,10 @@ def grad(self, grad0, _, grad1):
 class TestFunctionWithEagerExecution(TestFunction):
     @classmethod
     def setUpClass(self):
-        jt.flags.eager_execution = 1
+        jt.flags.lazy_execution = 0
     @classmethod
     def tearDownClass(self):
-        jt.flags.eager_execution = 0
+        jt.flags.lazy_execution = 1
 
 if __name__ == "__main__":
     unittest.main()

python/jittor/test/test_matmul.py

@@ -341,6 +341,7 @@ def check(a_shape, b_shape):
         check([8,1,3,4], [10,4,5])
         check([5,10,3,4], [5,10,4,5])
 
+    @unittest.skipIf(not jt.compiler.has_cuda, "No CUDA found")
     @jt.flag_scope(use_cuda=1)
     def test_matmul_example2_cuda(self):
         self.test_matmul_example2()

python/jittor/test/test_numpy_code_op.py

@@ -58,9 +58,9 @@ def check():
             one=numpy.ones(a.shape)
             assert numpy.allclose(da.data,one*2.0)
 
-        jt.flags.use_cuda = 0
-        check()
-        jt.flags.use_cuda = 1
+        if jt.has_cuda:
+            with jt.flag_scope(use_cuda=1):
+                check()
         check()
 
     def test(self):
@@ -92,9 +92,9 @@ def check():
             one=numpy.ones(a.shape)
             assert numpy.allclose(da.data,one*2.0)
 
-        jt.flags.use_cuda = 0
-        check()
-        jt.flags.use_cuda = 1
+        if jt.has_cuda:
+            with jt.flag_scope(use_cuda=1):
+                check()
         check()
 
     def test_multi_input(self):
@@ -139,9 +139,9 @@ def check():
             assert numpy.allclose(dda.data,one)
             assert numpy.allclose(ddb.data,mone)
 
-        jt.flags.use_cuda = 0
-        check()
-        jt.flags.use_cuda = 1
+        if jt.has_cuda:
+            with jt.flag_scope(use_cuda=1):
+                check()
         check()
 
     @unittest.skipIf(True, "Memory leak testing is not in progress, Skip")

python/jittor/test/test_parallel_pass.py

@@ -36,7 +36,7 @@ def check(self, use_int32):
         b = jt.random((n, n))
         a.data, b.data
         with jt.profile_scope(compile_options = {
-            "compile_shapes":1, "parallel":1
+            "compile_shapes":1, "parallel":2, "try_use_32bit_index":use_int32
         }, try_use_32bit_index = use_int32) as rep:
             c = a + b
             nc = c.data

src/utils/log.cc

@@ -335,7 +335,7 @@ void test_log_time(std::ostream* out) {
         auto finish = std::chrono::high_resolution_clock::now();
         auto total_ns =  std::chrono::duration_cast<std::chrono::nanoseconds>(finish-start).count();
         LOGi << "total_ns" << total_ns << "each_ns" << total_ns/n;
-        CHECKop(total_ns/n,<=,6000);
+        CHECKop(total_ns/n,<=,6500);
     };
     std::list<std::thread> ts;
     for (int i=0; i<nthread; i++) ts.emplace_back(log_lot);

src/var_holder.cc

@@ -16,14 +16,14 @@
 
 namespace jittor {
 
-DEFINE_FLAG(int, eager_execution, 0, "Use Eager execution rather than lazy execution, This flag makes error message and traceback infomation better. But this flag will raise memory consumption and lower the performance.");
+DEFINE_FLAG(int, lazy_execution, 1, "Default enabled, if disable, use immediately eager execution rather than lazy execution, This flag makes error message and traceback infomation better. But this flag will raise memory consumption and lower the performance.");
 
 list<VarHolder*> VarHolder::hold_vars;
 
 void add_hold_vars(VarHolder* self) {
     VarHolder::hold_vars.push_front(self);
     self->iter = VarHolder::hold_vars.begin();
-    if (!eager_execution) return;
+    if (lazy_execution) return;
     auto v = self->var;
     for (int i=0; i<5; i++) {
         auto op = v->input();