[kernels/quantized] Enable -Werror and fix format specifier warnings

cbilgin · cbilgin · commit e20f5e0ca6be · 2026-03-10T16:57:24.000-07:00
Summary:
Enable -Werror for kernels/quantized and fix format specifier mismatches
that were causing -Wformat warnings when tensor dimension/size values
(int64_t in ATen mode) were formatted with %zd (for ssize_t).

Changes:
- Add -Werror to _common_compile_options in kernels/quantized/CMakeLists.txt
- Replace %zd with ET_PRI_TENSOR_DIM for tensor .dim(), .size(), and .numel() calls in:
  - op_embedding.cpp
  - embeddingxb.cpp
  - op_choose_qparams.cpp
  - op_quantize.cpp
- Fix ET_CHECK_VALID_DIM macro in tensor_util.h to use ET_PRI_TENSOR_DIM

ET_PRI_TENSOR_DIM is the portable format specifier macro that adapts to
the build mode (PRId64 for ATen mode, "zd" for portable mode).

Test Plan:
Build with: cmake --build cmake-out -j9 --target quantized_kernels
Verify no -Wformat errors with -Werror enabled.
diff --git a/kernels/quantized/CMakeLists.txt b/kernels/quantized/CMakeLists.txt
@@ -21,7 +21,7 @@ if(NOT EXECUTORCH_ROOT)
   set(EXECUTORCH_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/../..)
 endif()
 
-set(_common_compile_options -Wno-deprecated-declarations)
+set(_common_compile_options -Wno-deprecated-declarations -Werror)
 
 include(${EXECUTORCH_ROOT}/tools/cmake/Utils.cmake)
 include(${EXECUTORCH_ROOT}/tools/cmake/Codegen.cmake)
diff --git a/kernels/quantized/cpu/embeddingxb.cpp b/kernels/quantized/cpu/embeddingxb.cpp
@@ -72,17 +72,17 @@ void check_embedding_xbit_args(
   ET_CHECK_MSG(8 % weight_nbit == 0, "nbit must divide 8");
 
   ET_CHECK_MSG(
-      weight.dim() == 2, "weight must be 2D but got() %zd dims", weight.dim());
+      weight.dim() == 2, "weight must be 2D but got() %" ET_PRI_TENSOR_DIM " dims", weight.dim());
 
   ET_CHECK_MSG(
       weight_scales.dim() == 1 || weight_scales.dim() == 2,
-      "weight_scales must be 1D or 2D but got() %zd dims",
+      "weight_scales must be 1D or 2D but got() %" ET_PRI_TENSOR_DIM " dims",
       weight_scales.dim());
 
   ET_CHECK_MSG(
       weight_scales.size(0) == weight.size(0),
-      "Number of scales must be == weight.size(0)=%zd"
-      ", but got %zd",
+      "Number of scales must be == weight.size(0)=%" ET_PRI_TENSOR_DIM
+      ", but got %" ET_PRI_TENSOR_DIM,
       weight_scales.size(0),
       weight.size(0));
 
@@ -91,8 +91,8 @@ void check_embedding_xbit_args(
     ET_CHECK_MSG(
         // each 8b uint8 column is packed_values_per_byte columns
         get_embedding_dim(weight.size(1), weight_nbit) % num_groups == 0,
-        "Number of groups must divide weight.size(1)=%zd"
-        ", but got # of groups = %zd",
+        "Number of groups must divide weight.size(1)=%" ET_PRI_TENSOR_DIM
+        ", but got # of groups = %" ET_PRI_TENSOR_DIM,
         weight.size(1),
         num_groups);
   }
@@ -132,8 +132,8 @@ void check_embedding_xbit_args(
       ET_CHECK_MSG(
           opt_weight_zero_points.value().size(i) == weight_scales.size(i),
           "Dimension size misatch at dim %" PRIi32
-          "Weight_zero_point size = %zd"
-          ", weight_scales size = %zd.",
+          " Weight_zero_point size = %" ET_PRI_TENSOR_DIM
+          ", weight_scales size = %" ET_PRI_TENSOR_DIM ".",
           i,
           opt_weight_zero_points.value().size(i),
           weight_scales.size(i));
diff --git a/kernels/quantized/cpu/op_choose_qparams.cpp b/kernels/quantized/cpu/op_choose_qparams.cpp
@@ -63,23 +63,23 @@ void check_quantize_per_tensor_args(
     for (auto i = 0; i < input.dim() - 1; i++) {
       ET_CHECK_MSG(
           scale_out.size(i) == input.size(i),
-          "Exepcted scale to have the same number of elements at dimentions %d got %zd",
+          "Exepcted scale to have the same number of elements at dimentions %d got %" ET_PRI_TENSOR_DIM,
           i,
           scale_out.size(i));
       ET_CHECK_MSG(
           zero_point_out.size(i) == input.size(i),
-          "Exepcted zero pont to have the same number of elements at dimentions %d got %zd",
+          "Exepcted zero pont to have the same number of elements at dimentions %d got %" ET_PRI_TENSOR_DIM,
           i,
           zero_point_out.size(i));
     }
     ET_CHECK_MSG(
         scale_out.size(input.dim() - 1) == 1,
-        "Exepcted scale to have only one element at dimentions %zd but got %zd",
+        "Exepcted scale to have only one element at dimentions %" ET_PRI_TENSOR_DIM " but got %" ET_PRI_TENSOR_DIM,
         input.dim() - 1,
         scale_out.size(input.dim() - 1));
     ET_CHECK_MSG(
         zero_point_out.size(input.dim() - 1) == 1,
-        "Exepcted zero point to have only one element at dimentions %zd but got %zd",
+        "Exepcted zero point to have only one element at dimentions %" ET_PRI_TENSOR_DIM " but got %" ET_PRI_TENSOR_DIM,
         input.dim() - 1,
         zero_point_out.size(input.dim() - 1));
   } else {
diff --git a/kernels/quantized/cpu/op_embedding.cpp b/kernels/quantized/cpu/op_embedding.cpp
@@ -34,26 +34,26 @@ void check_embedding_byte_args(
     std::optional<ScalarType> out_dtype,
     Tensor& out) {
   ET_CHECK_MSG(
-      weight.dim() == 2, "weight must be 2D but got() %zd dims", weight.dim());
+      weight.dim() == 2, "weight must be 2D but got() %" ET_PRI_TENSOR_DIM " dims", weight.dim());
 
   ET_CHECK_MSG(
       weight_scales.dim() == 1 || weight_scales.dim() == 2,
-      "weight_scales must be 1D or 2D but got() %zd dims",
+      "weight_scales must be 1D or 2D but got() %" ET_PRI_TENSOR_DIM " dims",
       weight_scales.dim());
 
   ET_CHECK_MSG(
       weight_scales.size(0) == weight.size(0),
-      "Number of scales must be == weight.size(0)=%zd"
-      ", but got %zd",
+      "Number of scales must be == weight.size(0)=%" ET_PRI_TENSOR_DIM
+      ", but got %" ET_PRI_TENSOR_DIM,
       weight_scales.size(0),
       weight.size(0));
 
   if (weight_scales.dim() == 2) {
     auto num_groups = weight_scales.size(1);
     ET_CHECK_MSG(
         weight.size(1) % num_groups == 0,
-        "Number of groups must divide weight.size(1)=%zd"
-        ", but got # of groups = %zd",
+        "Number of groups must divide weight.size(1)=%" ET_PRI_TENSOR_DIM
+        ", but got # of groups = %" ET_PRI_TENSOR_DIM,
         weight.size(1),
         num_groups);
   }
@@ -94,8 +94,8 @@ void check_embedding_byte_args(
       ET_CHECK_MSG(
           opt_weight_zero_points.value().size(i) == weight_scales.size(i),
           "Dimension size misatch at dim %" PRIi32
-          "Weight_zero_point size = %zd"
-          ", weight_scales size = %zd.",
+          " Weight_zero_point size = %" ET_PRI_TENSOR_DIM
+          ", weight_scales size = %" ET_PRI_TENSOR_DIM ".",
           i,
           opt_weight_zero_points.value().size(i),
           weight_scales.size(i));
@@ -158,14 +158,14 @@ void embedding_byte_per_channel(
     ET_CHECK_MSG(
         index >= 0 && index < weight.size(0),
         "Index out of bounds for weight: index %" PRId64
-        " must be in range [0, %zd)",
+        " must be in range [0, %" ET_PRI_TENSOR_DIM ")",
         index,
         weight.size(0));
 
     ET_CHECK_MSG(
         index >= 0 && index < weight_scales.size(0),
         "Index out of bounds for weight_scales: index %" PRId64
-        " must be in range [0, %zd)",
+        " must be in range [0, %" ET_PRI_TENSOR_DIM ")",
         index,
         weight_scales.size(0));
 
diff --git a/kernels/quantized/cpu/op_quantize.cpp b/kernels/quantized/cpu/op_quantize.cpp
@@ -431,7 +431,7 @@ Tensor& quantize_per_channel_out(
 
   ET_CHECK_MSG(
       scale.numel() == input.size(axis),
-      "scale.numel() %zd != input.size(axis) %zd",
+      "scale.numel() %" ET_PRI_TENSOR_DIM " != input.size(axis) %" ET_PRI_TENSOR_DIM,
       scale.numel(),
       input.size(axis));
 
@@ -442,7 +442,7 @@ Tensor& quantize_per_channel_out(
 
   ET_CHECK_MSG(
       zero_point.numel() == input.size(axis),
-      "zero_point.numel() %zd != input.size(axis) %zd",
+      "zero_point.numel() %" ET_PRI_TENSOR_DIM " != input.size(axis) %" ET_PRI_TENSOR_DIM,
       zero_point.numel(),
       input.size(axis));
 
diff --git a/runtime/core/exec_aten/util/tensor_util.h b/runtime/core/exec_aten/util/tensor_util.h
@@ -44,13 +44,13 @@
       UPPER_BOUND,                                          \
       UPPER_BOUND)
 
-#define ET_CHECK_VALID_DIM(DIM, UPPER_BOUND)              \
-  ET_CHECK_MSG(                                           \
-      DIM >= -static_cast<int64_t>(UPPER_BOUND) &&        \
-          DIM < static_cast<int64_t>(UPPER_BOUND),        \
-      "dim %" PRId64 " must be within range [-%zd, %zd)", \
-      DIM,                                                \
-      UPPER_BOUND,                                        \
+#define ET_CHECK_VALID_DIM(DIM, UPPER_BOUND)                                  \
+  ET_CHECK_MSG(                                                               \
+      DIM >= -static_cast<int64_t>(UPPER_BOUND) &&                            \
+          DIM < static_cast<int64_t>(UPPER_BOUND),                            \
+      "dim %" PRId64 " must be within range [-%" ET_PRI_TENSOR_DIM ", %" ET_PRI_TENSOR_DIM ")", \
+      DIM,                                                                    \
+      UPPER_BOUND,                                                            \
       UPPER_BOUND)
 
 #define ET_CHECK_NON_ZERO_DIM_SIZE(DIM, T)           \
