Update refmodel apply_scale_32: adjust range checking

Fix up generated values for rescale tests

Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I28fc3b8f189d25a7ad8e5172d4d8a43b86820fcf

2 files changed, 32 insertions, 0 deletions

diff --git a/reference_model/src/quant_util.h b/reference_model/src/quant_util.h
index 4f6a525..8c1b391 100644
--- a/reference_model/src/quant_util.h
+++ b/reference_model/src/quant_util.h
@@ -55,6 +55,16 @@ public:
                 "apply_scale_32(): shift value should stay within [2, 62] but is " + std::to_string(shift);
             throw desc;
         }
+        int64_t low_val = -1L << (shift-2);
+        int64_t high_val = 1L << (shift-2);
+        if (value < low_val || value >= high_val)
+        {
+            std::string desc =
+                "apply_scale_32(): value should stay within [" +
+                std::to_string(low_val) + ", " + std::to_string(high_val-1) +
+                "] but is " + std::to_string(value) + " using shift of " + std::to_string(shift);
+            throw desc;
+        }
         int64_t round = 1L << (shift - 1);
         if (double_round)
         {
diff --git a/verif/generator/tosa_test_gen.py b/verif/generator/tosa_test_gen.py
index 7e4cb1d..e60e643 100644
--- a/verif/generator/tosa_test_gen.py
+++ b/verif/generator/tosa_test_gen.py
@@ -5135,13 +5135,35 @@ class TosaTestGen:
 
         multiplier_arr = np.int32(np.zeros(shape=[nc]))
         shift_arr = np.int32(np.zeros(shape=[nc]))
+        min_shift_value_arr = np.int64(np.zeros(shape=[nc]))
+        max_shift_value_arr = np.int64(np.zeros(shape=[nc]))
 
         for i in range(nc):
             multiplier_arr[i], shift_arr[i] = TosaQuantGen.computeMultiplierAndShift(
                 scale_arr[i], scale32
             )
+            min_shift_value_arr[i] = -1 << (shift_arr[i] - 2)
+            max_shift_value_arr[i] = (1 << (shift_arr[i] - 2)) - 1
 
         # print('multiplier {} shift {} inzp {} outzp {}'.format(multiplier_arr, shift_arr, input_zp, output_zp))
+        if error_name is None:
+            # Make sure random values are within speicification
+            # REQUIRES(value >= (-1<<(shift-2)) && value < (1<<(shift-2))
+            assert val.placeholderFilename
+            values = np.load(
+                os.path.join(self.basePath, self.testPath, val.placeholderFilename)
+            )
+            val_adj = np.subtract(values, input_zp)
+            val_adj = np.maximum(val_adj, min_shift_value_arr, dtype=values.dtype)
+            val_adj = np.minimum(val_adj, max_shift_value_arr, dtype=values.dtype)
+            val_adj = np.add(val_adj, input_zp)
+            if not np.all(np.array_equal(values, val_adj)):
+                # Values changed so overwrite file with new values
+                np.save(
+                    os.path.join(self.basePath, self.testPath, val.placeholderFilename),
+                    val_adj,
+                    False,
+                )
 
         # Invalidate Input/Output list for error if checks.
         input_list = [val.name]