Update on "[quant] Add reflection padding to conv"

Differential Revision: [D25394384](https://our.internmc.facebook.com/intern/diff/D25394384)

[ghstack-poisoned]

.circleci/cimodel/data/dimensions.py

@@ -8,7 +8,7 @@
 
 ROCM_VERSIONS = [
     "3.10",
-    "4.0",
+    "4.0.1",
 ]
 
 ROCM_VERSION_LABELS = ["rocm" + v for v in ROCM_VERSIONS]

.circleci/config.yml

@@ -2165,49 +2165,49 @@ workflows:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           docker_image: "pytorch/manylinux-rocm:3.10"
       - binary_linux_build:
-          name: binary_linux_manywheel_3_6m_rocm4_0_devtoolset7_nightly_build
-          build_environment: "manywheel 3.6m rocm4.0 devtoolset7"
+          name: binary_linux_manywheel_3_6m_rocm4_0_1_devtoolset7_nightly_build
+          build_environment: "manywheel 3.6m rocm4.0.1 devtoolset7"
           filters:
             branches:
               only:
                 - /.*/
             tags:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
-          docker_image: "pytorch/manylinux-rocm:4.0"
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
       - binary_linux_build:
-          name: binary_linux_manywheel_3_7m_rocm4_0_devtoolset7_nightly_build
-          build_environment: "manywheel 3.7m rocm4.0 devtoolset7"
+          name: binary_linux_manywheel_3_7m_rocm4_0_1_devtoolset7_nightly_build
+          build_environment: "manywheel 3.7m rocm4.0.1 devtoolset7"
           filters:
             branches:
               only:
                 - /.*/
             tags:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
-          docker_image: "pytorch/manylinux-rocm:4.0"
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
       - binary_linux_build:
-          name: binary_linux_manywheel_3_8m_rocm4_0_devtoolset7_nightly_build
-          build_environment: "manywheel 3.8m rocm4.0 devtoolset7"
+          name: binary_linux_manywheel_3_8m_rocm4_0_1_devtoolset7_nightly_build
+          build_environment: "manywheel 3.8m rocm4.0.1 devtoolset7"
           filters:
             branches:
               only:
                 - /.*/
             tags:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
-          docker_image: "pytorch/manylinux-rocm:4.0"
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
       - binary_linux_build:
-          name: binary_linux_manywheel_3_9m_rocm4_0_devtoolset7_nightly_build
-          build_environment: "manywheel 3.9m rocm4.0 devtoolset7"
+          name: binary_linux_manywheel_3_9m_rocm4_0_1_devtoolset7_nightly_build
+          build_environment: "manywheel 3.9m rocm4.0.1 devtoolset7"
           filters:
             branches:
               only:
                 - /.*/
             tags:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
-          docker_image: "pytorch/manylinux-rocm:4.0"
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
       - binary_linux_build:
           name: binary_linux_conda_3_6_cpu_devtoolset7_nightly_build
           build_environment: "conda 3.6 cpu devtoolset7"
@@ -3551,8 +3551,8 @@ workflows:
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - binary_linux_test:
-          name: binary_linux_manywheel_3_6m_rocm4_0_devtoolset7_nightly_test
-          build_environment: "manywheel 3.6m rocm4.0 devtoolset7"
+          name: binary_linux_manywheel_3_6m_rocm4_0_1_devtoolset7_nightly_test
+          build_environment: "manywheel 3.6m rocm4.0.1 devtoolset7"
           filters:
             branches:
               only:
@@ -3561,13 +3561,13 @@ workflows:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           requires:
-            - binary_linux_manywheel_3_6m_rocm4_0_devtoolset7_nightly_build
-          docker_image: "pytorch/manylinux-rocm:4.0"
+            - binary_linux_manywheel_3_6m_rocm4_0_1_devtoolset7_nightly_build
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - binary_linux_test:
-          name: binary_linux_manywheel_3_7m_rocm4_0_devtoolset7_nightly_test
-          build_environment: "manywheel 3.7m rocm4.0 devtoolset7"
+          name: binary_linux_manywheel_3_7m_rocm4_0_1_devtoolset7_nightly_test
+          build_environment: "manywheel 3.7m rocm4.0.1 devtoolset7"
           filters:
             branches:
               only:
@@ -3576,13 +3576,13 @@ workflows:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           requires:
-            - binary_linux_manywheel_3_7m_rocm4_0_devtoolset7_nightly_build
-          docker_image: "pytorch/manylinux-rocm:4.0"
+            - binary_linux_manywheel_3_7m_rocm4_0_1_devtoolset7_nightly_build
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - binary_linux_test:
-          name: binary_linux_manywheel_3_8m_rocm4_0_devtoolset7_nightly_test
-          build_environment: "manywheel 3.8m rocm4.0 devtoolset7"
+          name: binary_linux_manywheel_3_8m_rocm4_0_1_devtoolset7_nightly_test
+          build_environment: "manywheel 3.8m rocm4.0.1 devtoolset7"
           filters:
             branches:
               only:
@@ -3591,13 +3591,13 @@ workflows:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           requires:
-            - binary_linux_manywheel_3_8m_rocm4_0_devtoolset7_nightly_build
-          docker_image: "pytorch/manylinux-rocm:4.0"
+            - binary_linux_manywheel_3_8m_rocm4_0_1_devtoolset7_nightly_build
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - binary_linux_test:
-          name: binary_linux_manywheel_3_9m_rocm4_0_devtoolset7_nightly_test
-          build_environment: "manywheel 3.9m rocm4.0 devtoolset7"
+          name: binary_linux_manywheel_3_9m_rocm4_0_1_devtoolset7_nightly_test
+          build_environment: "manywheel 3.9m rocm4.0.1 devtoolset7"
           filters:
             branches:
               only:
@@ -3606,8 +3606,8 @@ workflows:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           requires:
-            - binary_linux_manywheel_3_9m_rocm4_0_devtoolset7_nightly_build
-          docker_image: "pytorch/manylinux-rocm:4.0"
+            - binary_linux_manywheel_3_9m_rocm4_0_1_devtoolset7_nightly_build
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - binary_linux_test:
@@ -5129,10 +5129,10 @@ workflows:
           package_type: manywheel
           upload_subfolder: rocm3.10
       - binary_upload:
-          name: binary_linux_manywheel_3_6m_rocm4_0_devtoolset7_nightly_upload
+          name: binary_linux_manywheel_3_6m_rocm4_0_1_devtoolset7_nightly_upload
           context: org-member
           requires:
-            - binary_linux_manywheel_3_6m_rocm4_0_devtoolset7_nightly_test
+            - binary_linux_manywheel_3_6m_rocm4_0_1_devtoolset7_nightly_test
           filters:
             branches:
               only:
@@ -5141,12 +5141,12 @@ workflows:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           package_type: manywheel
-          upload_subfolder: rocm4.0
+          upload_subfolder: rocm4.0.1
       - binary_upload:
-          name: binary_linux_manywheel_3_7m_rocm4_0_devtoolset7_nightly_upload
+          name: binary_linux_manywheel_3_7m_rocm4_0_1_devtoolset7_nightly_upload
           context: org-member
           requires:
-            - binary_linux_manywheel_3_7m_rocm4_0_devtoolset7_nightly_test
+            - binary_linux_manywheel_3_7m_rocm4_0_1_devtoolset7_nightly_test
           filters:
             branches:
               only:
@@ -5155,12 +5155,12 @@ workflows:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           package_type: manywheel
-          upload_subfolder: rocm4.0
+          upload_subfolder: rocm4.0.1
       - binary_upload:
-          name: binary_linux_manywheel_3_8m_rocm4_0_devtoolset7_nightly_upload
+          name: binary_linux_manywheel_3_8m_rocm4_0_1_devtoolset7_nightly_upload
           context: org-member
           requires:
-            - binary_linux_manywheel_3_8m_rocm4_0_devtoolset7_nightly_test
+            - binary_linux_manywheel_3_8m_rocm4_0_1_devtoolset7_nightly_test
           filters:
             branches:
               only:
@@ -5169,12 +5169,12 @@ workflows:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           package_type: manywheel
-          upload_subfolder: rocm4.0
+          upload_subfolder: rocm4.0.1
       - binary_upload:
-          name: binary_linux_manywheel_3_9m_rocm4_0_devtoolset7_nightly_upload
+          name: binary_linux_manywheel_3_9m_rocm4_0_1_devtoolset7_nightly_upload
           context: org-member
           requires:
-            - binary_linux_manywheel_3_9m_rocm4_0_devtoolset7_nightly_test
+            - binary_linux_manywheel_3_9m_rocm4_0_1_devtoolset7_nightly_test
           filters:
             branches:
               only:
@@ -5183,7 +5183,7 @@ workflows:
               only:
                 - /v[0-9]+(\.[0-9]+)*-rc[0-9]+/
           package_type: manywheel
-          upload_subfolder: rocm4.0
+          upload_subfolder: rocm4.0.1
       - binary_upload:
           name: binary_linux_conda_3_6_cpu_devtoolset7_nightly_upload
           context: org-member
@@ -7801,51 +7801,51 @@ workflows:
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - smoke_linux_test:
-          name: smoke_linux_manywheel_3_6m_rocm4_0_devtoolset7_nightly
-          build_environment: "manywheel 3.6m rocm4.0 devtoolset7"
+          name: smoke_linux_manywheel_3_6m_rocm4_0_1_devtoolset7_nightly
+          build_environment: "manywheel 3.6m rocm4.0.1 devtoolset7"
           requires:
             - update_s3_htmls
           filters:
             branches:
               only:
                 - postnightly
-          docker_image: "pytorch/manylinux-rocm:4.0"
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - smoke_linux_test:
-          name: smoke_linux_manywheel_3_7m_rocm4_0_devtoolset7_nightly
-          build_environment: "manywheel 3.7m rocm4.0 devtoolset7"
+          name: smoke_linux_manywheel_3_7m_rocm4_0_1_devtoolset7_nightly
+          build_environment: "manywheel 3.7m rocm4.0.1 devtoolset7"
           requires:
             - update_s3_htmls
           filters:
             branches:
               only:
                 - postnightly
-          docker_image: "pytorch/manylinux-rocm:4.0"
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - smoke_linux_test:
-          name: smoke_linux_manywheel_3_8m_rocm4_0_devtoolset7_nightly
-          build_environment: "manywheel 3.8m rocm4.0 devtoolset7"
+          name: smoke_linux_manywheel_3_8m_rocm4_0_1_devtoolset7_nightly
+          build_environment: "manywheel 3.8m rocm4.0.1 devtoolset7"
           requires:
             - update_s3_htmls
           filters:
             branches:
               only:
                 - postnightly
-          docker_image: "pytorch/manylinux-rocm:4.0"
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - smoke_linux_test:
-          name: smoke_linux_manywheel_3_9m_rocm4_0_devtoolset7_nightly
-          build_environment: "manywheel 3.9m rocm4.0 devtoolset7"
+          name: smoke_linux_manywheel_3_9m_rocm4_0_1_devtoolset7_nightly
+          build_environment: "manywheel 3.9m rocm4.0.1 devtoolset7"
           requires:
             - update_s3_htmls
           filters:
             branches:
               only:
                 - postnightly
-          docker_image: "pytorch/manylinux-rocm:4.0"
+          docker_image: "pytorch/manylinux-rocm:4.0.1"
           use_cuda_docker_runtime: "1"
           resource_class: gpu.medium
       - smoke_linux_test:

aten/src/ATen/Utils.cpp

@@ -27,9 +27,8 @@ Tensor empty_cpu(
     c10::optional<Device> device_opt,
     c10::optional<bool> pin_memory_opt,
     c10::optional<c10::MemoryFormat> memory_format_opt) {
-  Device device = device_or_default(device_opt);
+  TORCH_INTERNAL_ASSERT_DEBUG_ONLY(device_or_default(device_opt).type() == DeviceType::CPU);
 
-  TORCH_CHECK(device.type() == DeviceType::CPU);
   check_size_nonnegative(size);
 
   bool pin_memory = pinned_memory_or_default(pin_memory_opt);

aten/src/ATen/core/aten_interned_strings.h

@@ -289,6 +289,7 @@ _(aten, diag_embed) \
 _(aten, diagflat) \
 _(aten, diagonal) \
 _(aten, fill_diagonal_) \
+_(aten, diff) \
 _(aten, digamma) \
 _(aten, dim) \
 _(aten, dist) \

aten/src/ATen/core/op_registration/op_whitelist.h

@@ -10,7 +10,7 @@
  * set of operators (+ dependencies) included.
  *
  * - Build with -DTORCH_OPERATOR_WHITELIST="aten::add;aten::sub" and only these
- *   two ops will be included in your build.  The whitelist records operators
+ *   two ops will be included in your build.  The allowlist records operators
  *   only, no overloads; if you include aten::add, all overloads of aten::add
  *   will be included.
  *
@@ -19,9 +19,9 @@
  * operator functions that weren't registered.
  * See Note [Selective build] for more details
  *
- * WARNING: The whitelist mechanism doesn't work for all ways you could go about
+ * WARNING: The allowlist mechanism doesn't work for all ways you could go about
  * registering an operator.  If the dispatch key / operator name is not
- * sufficiently obvious at compile time, then the whitelisting mechanism
+ * sufficiently obvious at compile time, then the allowlisting mechanism
  * will fail (and the operator will be included in the binary anyway).
  */
 
@@ -33,21 +33,21 @@ namespace c10 {
 
 namespace impl {
 
-// returns true iff whitelist contains item
+// returns true iff allowlist contains item
 // op_whitelist_contains("a;bc;d", "bc") == true
-constexpr bool op_whitelist_contains(string_view whitelist, string_view item) {
+constexpr bool op_whitelist_contains(string_view allowlist, string_view item) {
     //Choose a really big value for next so that if something goes wrong
     //this code will blow up in a hopefully detectable way.
     size_t next = std::numeric_limits<size_t>::max();
-    for (size_t cur = 0; cur <= whitelist.size(); cur = next) {
-      next = whitelist.find(';', cur);
+    for (size_t cur = 0; cur <= allowlist.size(); cur = next) {
+      next = allowlist.find(';', cur);
       if (next != string_view::npos) {
-        if (whitelist.substr(cur, next - cur).compare(item) == 0) {
+        if (allowlist.substr(cur, next - cur).compare(item) == 0) {
           return true;
         }
         next++;
       } else {
-        if (whitelist.substr(cur).compare(item) == 0) {
+        if (allowlist.substr(cur).compare(item) == 0) {
           return true;
         }
         break;
@@ -56,7 +56,7 @@ constexpr bool op_whitelist_contains(string_view whitelist, string_view item) {
     return false;
 }
 
-// Returns true iff the given op name is on the whitelist
+// Returns true iff the given op name is on the allowlist
 // and should be registered
 constexpr bool op_whitelist_check(string_view op_name) {
   assert(op_name.find("::") != string_view::npos);
@@ -68,16 +68,16 @@ constexpr bool op_whitelist_check(string_view op_name) {
   return op_whitelist_contains(
     C10_STRINGIZE(TORCH_OPERATOR_WHITELIST),
     // This function is majorly used for mobile selective build with
-    // root operators, where the overload is included in the whitelist.
+    // root operators, where the overload is included in the allowlist.
     op_name);
-    // // Strip overload name (as whitelist doesn't contain overloads)
+    // // Strip overload name (as allowlist doesn't contain overloads)
     // // Another function based on this may be added when there's usage
     // // on op names without overload.
     // OperatorNameView::parse(op_name).name);
 #endif
 }
 
-// Returns true iff the given schema string is on the whitelist
+// Returns true iff the given schema string is on the allowlist
 // and should be registered
 constexpr bool schema_whitelist_check(string_view schema) {
 #if defined(TORCH_FORCE_SCHEMA_REGISTRATION)
@@ -88,12 +88,12 @@ constexpr bool schema_whitelist_check(string_view schema) {
 }
 
 // schema_whitelist_check() implicitly depends on a macro, TORCH_OPERATOR_WHITELIST.
-// Add this API to pass arbitrary whitelist.
-constexpr bool op_whitelist_contains_name_in_schema(string_view whitelist, string_view schema) {
-  return op_whitelist_contains(whitelist, schema.substr(0, schema.find("(")));
+// Add this API to pass arbitrary allowlist.
+constexpr bool op_whitelist_contains_name_in_schema(string_view allowlist, string_view schema) {
+  return op_whitelist_contains(allowlist, schema.substr(0, schema.find("(")));
 }
 
-// Returns true iff the given dispatch key is on the whitelist
+// Returns true iff the given dispatch key is on the allowlist
 // and should be registered.  When we turn this on, the list of valid
 // mobile dispatch keys is hard coded (but you need to make sure
 // that you have the correct set of dispatch keys for this).