Added AVR asm version of ease8InOutQuad which is a little faster, and a little smaller than the C version; it also preserves one additional low-order bit of precision so while it may not give exactly the same numeric results as the C in all cases, it presumably gives 'better' (smoother) ones.

1 files changed, 30 insertions, 0 deletions

diff --git a/lib8tion.h b/lib8tion.h
index 37c0cba5..c982732e 100644
--- a/lib8tion.h
+++ b/lib8tion.h
@@ -576,6 +576,7 @@ LIB8STATIC uint8_t map8( uint8_t in, uint8_t rangeStart, uint8_t rangeEnd)
 
 /// ease8InOutQuad: 8-bit quadratic ease-in / ease-out function
 ///                Takes around 13 cycles on AVR
+#if EASE8_C == 1
 LIB8STATIC uint8_t ease8InOutQuad( uint8_t i)
 {
     uint8_t j = i;
@@ -590,6 +591,35 @@ LIB8STATIC uint8_t ease8InOutQuad( uint8_t i)
     return jj2;
 }
 
+#elif EASE8_AVRASM == 1
+// This AVR asm version of ease8InOutQuad preserves one more
+// low-bit of precision than the C version, and is also slightly
+// smaller and faster.
+LIB8STATIC uint8_t ease8InOutQuad(uint8_t val) {
+    uint8_t j=val;
+    asm volatile (
+      "sbrc %[val], 7 \n"
+      "com %[j]       \n"
+      "mul %[j], %[j] \n"
+      "add r0, %[j]   \n"
+      "ldi %[j], 0    \n"
+      "adc %[j], r1   \n"
+      "lsl r0         \n" // carry = high bit of low byte of mul product
+      "rol %[j]       \n" // j = (j * 2) + carry // preserve add'l bit of precision
+      "sbrc %[val], 7 \n"
+      "com %[j]       \n"
+      "clr __zero_reg__   \n"
+      : [j] "+a" (j)
+      : [val] "a" (val)
+      : "r0", "r1"
+      );
+    return j;
+}
+
+#else
+#error "No implementation for ease8InOutQuad available."
+#endif
+
 
 /// ease8InOutCubic: 8-bit cubic ease-in / ease-out function
 ///                 Takes around 18 cycles on AVR