diff --git a/library/bn_mul.h b/library/bn_mul.h
index 34719b9..6b8106f 100644
--- a/library/bn_mul.h
+++ b/library/bn_mul.h
@@ -743,10 +743,10 @@
 
 #define MULADDC_X1_CORE                                         \
            ".p2align  2                                 \n\t"   \
-            "ldr.w    %[a], [%[in]], #4                 \n\t"   \
-            "ldr.w    %[b], [%[acc]]                    \n\t"   \
+            "ldr      %[a], [%[in]], #4                 \n\t"   \
+            "ldr      %[b], [%[acc]]                    \n\t"   \
             "umaal    %[b], %[carry], %[scalar], %[a]   \n\t"   \
-            "str.w    %[b], [%[acc]], #4                \n\t"
+            "str      %[b], [%[acc]], #4                \n\t"
 
 #define MULADDC_X1_STOP                                      \
             : [a]      "=&r" (tmp_a),                        \
@@ -777,14 +777,14 @@
              *   2 cycles, while subsequent loads/stores are single-cycle. */
 #define MULADDC_X2_CORE                                           \
            ".p2align  2                                   \n\t"   \
-            "ldr.w    %[a0], [%[in]],  #+8                \n\t"   \
-            "ldr.w    %[b0], [%[acc]], #+8                \n\t"   \
-            "ldr.w    %[a1], [%[in],  #-4]                \n\t"   \
-            "ldr.w    %[b1], [%[acc], #-4]                \n\t"   \
+            "ldr      %[a0], [%[in]],  #+8                \n\t"   \
+            "ldr      %[b0], [%[acc]], #+8                \n\t"   \
+            "ldr      %[a1], [%[in],  #-4]                \n\t"   \
+            "ldr      %[b1], [%[acc], #-4]                \n\t"   \
             "umaal    %[b0], %[carry], %[scalar], %[a0]   \n\t"   \
             "umaal    %[b1], %[carry], %[scalar], %[a1]   \n\t"   \
-            "str.w    %[b0], [%[acc], #-8]                \n\t"   \
-            "str.w    %[b1], [%[acc], #-4]                \n\t"
+            "str      %[b0], [%[acc], #-8]                \n\t"   \
+            "str      %[b1], [%[acc], #-4]                \n\t"
 
 #define MULADDC_X2_STOP                                      \
             : [a0]     "=&r" (tmp_a0),                       \
diff --git a/tests/scripts/all.sh b/tests/scripts/all.sh
index e5d3afc..7915dfb 100755
--- a/tests/scripts/all.sh
+++ b/tests/scripts/all.sh
@@ -2948,6 +2948,7 @@
     scripts/config.py baremetal
     # armc[56] don't support SHA-512 intrinsics
     scripts/config.py unset MBEDTLS_SHA512_USE_A64_CRYPTO_IF_PRESENT
+    scripts/config.py set MBEDTLS_HAVE_ASM
 
     make CC="$ARMC5_CC" AR="$ARMC5_AR" WARNING_CFLAGS='--strict --c99' lib
 
@@ -2956,20 +2957,25 @@
 
     make clean
 
+    # Compile with -O1 since some Arm inline assembly is disabled for -O0.
+
     # ARM Compiler 6 - Target ARMv7-A
-    armc6_build_test "--target=arm-arm-none-eabi -march=armv7-a"
+    armc6_build_test "-O1 --target=arm-arm-none-eabi -march=armv7-a"
 
     # ARM Compiler 6 - Target ARMv7-M
-    armc6_build_test "--target=arm-arm-none-eabi -march=armv7-m"
+    armc6_build_test "-O1 --target=arm-arm-none-eabi -march=armv7-m"
+
+    # ARM Compiler 6 - Target ARMv7-M+DSP
+    armc6_build_test "-O1 --target=arm-arm-none-eabi -march=armv7-m+dsp"
 
     # ARM Compiler 6 - Target ARMv8-A - AArch32
-    armc6_build_test "--target=arm-arm-none-eabi -march=armv8.2-a"
+    armc6_build_test "-O1 --target=arm-arm-none-eabi -march=armv8.2-a"
 
     # ARM Compiler 6 - Target ARMv8-M
-    armc6_build_test "--target=arm-arm-none-eabi -march=armv8-m.main"
+    armc6_build_test "-O1 --target=arm-arm-none-eabi -march=armv8-m.main"
 
     # ARM Compiler 6 - Target ARMv8.2-A - AArch64
-    armc6_build_test "--target=aarch64-arm-none-eabi -march=armv8.2-a+crypto"
+    armc6_build_test "-O1 --target=aarch64-arm-none-eabi -march=armv8.2-a+crypto"
 }
 
 component_test_tls13_only () {