jtagtap, swdptap, timing: fixup! Update the delay_busy, for-loop form

src/platforms/common/jtagtap.c

@@ -73,15 +73,8 @@ static void jtagtap_reset(void)
 /* Busy-looping delay snippet for GPIO bitbanging (rely on inlining) */
 static inline void platform_delay_busy(const uint32_t loops)
 {
-#if 1
-	register uint32_t i = loops;
-	do {
-		__asm__("nop");
-	} while (--i > 0U);
-#else
-	for (register uint32_t counter = loops; --counter > 0;)
-		__asm__("nop");
-#endif
+	for (register uint32_t counter = loops; --counter > 0U;)
+		__asm__("");
 }
 
 static bool jtagtap_next_clk_delay()

src/platforms/common/stm32/timing_stm32.c

@@ -216,6 +216,7 @@ uint32_t platform_max_frequency_get(void)
 void platform_delay_busy(const uint32_t loops)
 {
 	/* Avoid using `volatile` variables which incur stack accesses */
+#if 0
 	register uint32_t i = loops;
 	do {
 		/*
@@ -226,4 +227,13 @@ void platform_delay_busy(const uint32_t loops)
 		 */
 		__asm__("nop");
 	} while (--i > 0U);
+#else
+	/*
+	 * Another version which still assembles to SUBS+BNE.N; the NOP can be eliminated
+	 * in favor of an empty inline asm/volatile block if it's enough to suppress DCE.
+	 * Note that the predecrement has to be merged into the condition expression.
+	 */
+	for (register uint32_t i = loops; --i > 0U;)
+		__asm__("");
+#endif
 }

src/platforms/common/swdptap.c

@@ -57,15 +57,8 @@ void swdptap_init(void)
 /* Busy-looping delay snippet for GPIO bitbanging (rely on inlining) */
 static inline void platform_delay_busy(const uint32_t loops)
 {
-#if 1
-	register uint32_t i = loops;
-	do {
-		__asm__("nop");
-	} while (--i > 0U);
-#else
-	for (register uint32_t counter = loops; counter > 0; --counter)
-		__asm__("nop");
-#endif
+	for (register uint32_t counter = loops; --counter > 0U;)
+		__asm__("");
 }
 
 static void swdptap_turnaround(const swdio_status_t dir)