[cuda] Fix ti.random(dtype=ti.f64)

.github/pull_request_template.md

@@ -2,22 +2,19 @@ Related issue = #
 
 <!--
 Thanks for your PR!
-If it's your first time contributing to Taichi, please make sure you have read our Contributor Guideline:
-  https://taichi.readthedocs.io/en/latest/contributor_guide.html
+If it is your first time contributing to Taichi, please read our Contributor Guideline:
+  https://taichi.graphics/contribution/
 
-- Please always prepend your PR title with tags such as [CUDA], [Lang], [Doc], [Example], e.g.:
-    [Lang] Add ti.Complex as Taichi class
-- Use a lowercased tag for PRs that are invisible to end-users, i.e., won't be highlighted in changelog:
-    [cuda] [test] Fix out-of-memory error while running test
-- More details: http://taichi.readthedocs.io/en/latest/contributor_guide.html#prtags
+- Please always prepend your PR title with tags such as [CUDA], [Lang], [Doc], [Example]. E.g.:
+    [Lang] Add ti.sin
+- Use upper-case tags (e.g., [Metal]) for PRs that change public APIs. Otherwise, please use lower-case tags (e.g., [metal]).
+- More details: https://taichi.graphics/contribution/contributor_guide.html#pr-title-format-and-tags
 
-- Please fill the following blank with the issue number this PR related to (if any):
-    Related issue = #2345
-- If your PR will fix the issue **completely**, use the `close` or `fixes` keyword:
+- Please fill in the issue number that this PR relates to.
+- If your PR fixes the issue **completely**, use the `close` or `fixes` prefix so that GitHub automatically closes the issue when the PR is merged. For example,
     Related issue = close #2345
-- So that when the PR gets merged, GitHub will **automatically** close the issue #2345 for you.
-- If the PR doesn't belong to any existing issue, and it's a trivial change, feel free to leave it blank :)
-  -->
+- If the PR does not belong to any existing issue, free to leave it blank.
+-->
 
 [[Click here for the format server]](http://kun.csail.mit.edu:31415/)
 

taichi/runtime/llvm/runtime.cpp

@@ -1466,7 +1466,7 @@ f32 cuda_rand_f32(Context *context) {
   return cuda_rand_u32(context) * (1.0f / 4294967296.0f);
 }
 
-f32 cuda_rand_f64(Context *context) {
+f64 cuda_rand_f64(Context *context) {
   return cuda_rand_f32(context);
 }
 

tests/python/test_random.py

@@ -105,3 +105,12 @@ def gen(i: ti.i32):
         count += 1 if x[i] == x[i - 1] else 0
 
     assert count <= n * 0.15
+
+
+@ti.test(arch=[ti.cpu, ti.cuda])
+def test_random_f64():
+    @ti.kernel
+    def foo() -> ti.f64:
+        return ti.random(dtype=ti.f64)
+
+    foo()