⚡️ Optimize speed lookup for AVR

Co-Authored-By: tombrazier <[email protected]>

Author: thinkyhead

Marlin/src/HAL/AVR/math.h

@@ -84,32 +84,26 @@ FORCE_INLINE static uint16_t MultiU24X32toH16(uint32_t longIn1, uint32_t longIn2
   return intRes;
 }
 
-// intRes = intIn1 * intIn2 >> 8
+// charRes = charIn1 * charIn2 >> 8
 // uses:
-// r1, r0 for the result of mul. After the second mul, r0 holds bits 0-7 of the 24 bit result and
-// the top bit of r0 is used for rounding.
-// [tmp] to store 0.
-// [intRes] (A B) is bits 8-15 and is the returned value.
+// r1, r0 for the result of mul. After the mul, r0 holds bits 0-7 of the 16 bit result,
+//        and the top bit of r0 is used for rounding.
+// [charRes] is bits 8-15 and is the returned value.
 // [charIn1] is an 8 bit parameter.
-// [intIn2] (B A) is a 16 bit parameter.
+// [charIn2] is an 8 bit parameter.
 //
-FORCE_INLINE static uint16_t MultiU8X16toH16(uint8_t charIn1, uint16_t intIn2) {
-  uint8_t tmp;
-  uint16_t intRes;
+FORCE_INLINE static uint8_t MultiU8X8toH8(uint8_t charIn1, uint8_t charIn2) {
+  uint8_t charRes;
   __asm__ __volatile__ (
-    A("clr %[tmp]")
-    A("mul %[charIn1], %B[intIn2]")
-    A("movw %A[intRes], r0")
-    A("mul %[charIn1], %A[intIn2]")
-    A("lsl r0")
-    A("adc %A[intRes], r1")
-    A("adc %B[intRes], %[tmp]")
+    A("mul %[charIn1], %[charIn2]")
+    A("mov %[charRes], r1")
     A("clr r1")
-      : [intRes] "=&r" (intRes),
-        [tmp] "=&r" (tmp)
+    A("lsl r0")
+    A("adc %[charRes], r1")
+      : [charRes] "=&r" (charRes)
       : [charIn1] "d" (charIn1),
-        [intIn2] "d" (intIn2)
+        [charIn2] "d" (charIn2)
       : "cc"
   );
-  return intRes;
+  return charRes;
 }

Marlin/src/module/stepper.cpp

@@ -2069,22 +2069,21 @@ hal_timer_t Stepper::calc_timer_interval(uint32_t step_rate) {
 
     // AVR is able to keep up at 30khz Stepping ISR rate.
     constexpr uint32_t min_step_rate = (F_CPU) / 500000U; // i.e., 32 or 40
-    if (step_rate <= min_step_rate) { // lower step rates
-      step_rate = 0;
-      return uint16_t(pgm_read_word(uintptr_t(speed_lookuptable_slow)));
+    if (step_rate >= 0x0800) {  // higher step rate
+      const uintptr_t table_address = uintptr_t(&speed_lookuptable_fast[uint8_t(step_rate >> 8)]);
+      const uint16_t base = uint16_t(pgm_read_word(table_address));
+      const uint8_t gain = uint8_t(pgm_read_byte(table_address + 2));
+      return base - MultiU8X8toH8(uint8_t(step_rate & 0x00FF), gain);
     }
-    else {
+    else if (step_rate > min_step_rate) { // lower step rates
       step_rate -= min_step_rate; // Correct for minimal speed
-      if (step_rate >= 0x0800) {  // higher step rate
-        const uintptr_t table_address = uintptr_t(&speed_lookuptable_fast[uint8_t(step_rate >> 8)]);
-        const uint16_t gain = uint16_t(pgm_read_word(table_address + 2));
-        return uint16_t(pgm_read_word(table_address)) - MultiU8X16toH16(uint8_t(step_rate & 0x00FF), gain);
-      }
-      else { // lower step rates
-        const uintptr_t table_address = uintptr_t(&speed_lookuptable_slow[uint8_t(step_rate >> 3)]);
-        return uint16_t(pgm_read_word(table_address))
-               - ((uint16_t(pgm_read_word(table_address + 2)) * uint8_t(step_rate & 0x0007)) >> 3);
-      }
+      const uintptr_t table_address = uintptr_t(&speed_lookuptable_slow[uint8_t(step_rate >> 3)]);
+      return uint16_t(pgm_read_word(table_address))
+             - ((uint16_t(pgm_read_word(table_address + 2)) * uint8_t(step_rate & 0x0007)) >> 3);
+    }
+    else {
+      step_rate = 0;
+      return uint16_t(pgm_read_word(uintptr_t(speed_lookuptable_slow)));
     }
 
   #endif // !CPU_32_BIT