added exponent alignment routine, fix for FAC1 normalization

c64/basic/math/add_align_exponents.s

@@ -0,0 +1,169 @@
+// #LAYOUT# STD *       #TAKE
+// #LAYOUT# *   BASIC_0 #TAKE
+// #LAYOUT# *   *       #IGNORE
+
+//
+// Math package - align FAC1 and FAC2 exponents for addition
+//
+// See also:
+// - [CM64] Computes Mapping the Commodore 64 - page 115
+// - https://www.c64-wiki.com/wiki/BASIC-ROM
+
+// XXX finish and test this
+
+add_align_exponents:
+
+	// XXX assume here none of the exponents is 0, check if this is true for orignal ROMs
+
+	lda FAC1_exponent
+	cmp FAC2_exponent
+	beq add_align_exponents_done       // branch if exponents the same
+	bcs add_align_exponent_FAC2
+
+	// FALLTROUGH
+
+add_align_exponent_FAC1:
+
+	// First check the exponent difference
+
+	lda FAC2_exponent
+	sec
+	sbc FAC1_exponent
+
+	cmp #$28
+	bcs add_align_exponent_FAC1_zero   // branch if difference is very high
+
+	cmp #$08
+	bcc add_align_exponent_FAC1_bit
+	
+	// FALLTROUGH
+
+add_align_exponent_FAC1_byte:
+
+	// Add 8 to exponent, divide mantissa by 256
+
+	lda FAC1_exponent
+	clc
+	adc #$08
+	sta FAC1_exponent
+
+	lda FAC1_mantissa+3
+	sta FACOV
+	lda FAC1_mantissa+2
+	sta FAC1_mantissa+3
+	lda FAC1_mantissa+1
+	sta FAC1_mantissa+2
+	lda FAC1_mantissa+0
+	sta FAC1_mantissa+1
+	lda #$00
+	sta FAC1_mantissa+0
+
+	beq add_align_exponent_FAC1        // branch always - next iteration
+
+add_align_exponent_FAC1_bit:
+
+	// Increment exponent, divide mantissa by 2
+
+	inc FAC1_exponent
+
+	clc
+	ror FAC1_mantissa+0
+	ror FAC1_mantissa+1
+	ror FAC1_mantissa+2
+	ror FAC1_mantissa+3
+	ror FACOV
+
+	lda FAC1_exponent
+	cmp FAC2_exponent
+	bne add_align_exponent_FAC1_bit    // branch if next iteration needed
+
+	rts
+
+add_align_exponent_FAC1_zero:
+
+	lda #$00
+
+	sta FAC1_mantissa+0
+	sta FAC1_mantissa+1
+	sta FAC1_mantissa+2
+	sta FAC1_mantissa+3
+	sta FACOV
+
+	lda FAC2_exponent
+	sta FAC1_exponent
+
+	// FALLTROUGH
+
+add_align_exponents_done:
+
+	rts
+
+
+
+add_align_exponent_FAC2:
+
+	// First check the exponent difference, Carry already set here
+
+	lda FAC1_exponent
+	sbc FAC2_exponent
+
+	cmp #$20
+	bcs add_align_exponent_FAC2_zero   // branch if difference is very high
+
+	cmp #$08
+	bcc add_align_exponent_FAC2_bit
+	
+	// FALLTROUGH
+
+add_align_exponent_FAC2_byte:
+
+	// Add 8 to exponent, divide mantissa by 256
+
+	lda FAC2_exponent
+	clc
+	adc #$08
+	sta FAC2_exponent
+
+	lda FAC2_mantissa+2
+	sta FAC2_mantissa+3
+	lda FAC2_mantissa+1
+	sta FAC2_mantissa+2
+	lda FAC2_mantissa+0
+	sta FAC2_mantissa+1
+	lda #$00
+	sta FAC2_mantissa+0
+
+	beq add_align_exponent_FAC2        // branch always - next iteration
+
+add_align_exponent_FAC2_bit:
+
+	// Increment exponent, divide mantissa by 2
+
+	inc FAC2_exponent
+
+	clc
+	ror FAC2_mantissa+0
+	ror FAC2_mantissa+1
+	ror FAC2_mantissa+2
+	ror FAC2_mantissa+3
+
+	lda FAC1_exponent
+	cmp FAC2_exponent
+	bne add_align_exponent_FAC2_bit    // branch if next iteration needed
+
+	rts
+
+
+add_align_exponent_FAC2_zero:
+
+	lda #$00
+
+	sta FAC2_mantissa+0
+	sta FAC2_mantissa+1
+	sta FAC2_mantissa+2
+	sta FAC2_mantissa+3
+
+	lda FAC1_exponent
+	sta FAC2_exponent
+
+	rts

c64/basic/math/b862.add_align_exponent.jmp.s

@@ -0,0 +1,8 @@
+// #LAYOUT# STD *       #TAKE
+// #LAYOUT# *   BASIC_0 #TAKE
+// #LAYOUT# *   *       #IGNORE
+
+// Just a compatibility jump - our routine is longer than original one
+
+
+	jmp add_align_exponents

c64/basic/math/b862.interop

@@ -0,0 +1,3 @@
+This is an important floating point math routine, documented in many places.
+
+XXX - try to find real-life example

c64/basic/math/mov_FAC1_MEM.s

@@ -19,11 +19,8 @@
 
 mov_FAC1_MEM:
 
-	// Checked in the available documentation:
-	// - https://www.c64-wiki.com/wiki/Floating_point_arithmetic
-	// - https://www.lemon64.com/forum/viewtopic.php?t=67576&sid=d41e77cdddd28d6f9c9d52ce5b4e8dc3
-	// that $22/$23 (INDEX+0 and INDEX+1) are at least sometimes used for pointer storage.
-	// Confirmed by experimenting with original ROM that this is the case here too.
+	// Original routines use $22/$23 location for a vector too, see here:
+	// - https://codebase64.org/doku.php?id=base:kernal_floating_point_mathematics
 
 	sty INDEX+1
 	stx INDEX+0

c64/basic/math/mov_MEM_FAC1.s

@@ -23,11 +23,8 @@
 
 mov_MEM_FAC1:
 
-	// Checked in the available documentation:
-	// - https://www.c64-wiki.com/wiki/Floating_point_arithmetic
-	// - https://www.lemon64.com/forum/viewtopic.php?t=67576&sid=d41e77cdddd28d6f9c9d52ce5b4e8dc3
-	// that $22/$23 (INDEX+0 and INDEX+1) are at least sometimes used for pointer storage.
-	// Confirmed by experimenting with original ROM that this is the case here too.
+	// Original routines use $22/$23 location for a vector too, see here:
+	// - https://codebase64.org/doku.php?id=base:kernal_floating_point_mathematics
 
 	sty INDEX+1
 	sta INDEX+0

c64/basic/math/mov_MEM_FAC2.s

@@ -26,11 +26,8 @@
 
 mov_MEM_FAC2:
 
-	// Checked in the available documentation:
-	// - https://www.c64-wiki.com/wiki/Floating_point_arithmetic
-	// - https://www.lemon64.com/forum/viewtopic.php?t=67576&sid=d41e77cdddd28d6f9c9d52ce5b4e8dc3
-	// that $22/$23 (INDEX+0 and INDEX+1) are at least sometimes used for pointer storage.
-	// Confirmed by experimenting with original ROM that this is the case here too.
+	// Original routines use $22/$23 location for a vector too, see here:
+	// - https://codebase64.org/doku.php?id=base:kernal_floating_point_mathematics
 
 	sty INDEX+1
 	sta INDEX+0

c64/basic/math/normal_FAC1.s

@@ -16,30 +16,32 @@
 normal_FAC1:
 
 	lda FAC1_exponent
-	beq normal_FAC1_end                // branch if our float is 0
+	beq normal_FAC1_end                     // branch if our float is 0
 	
-	lda FAC1_mantissa+3
-	beq normal_FAC1_by_byte            // branch if we can move the whole byte
+	// FALLTROUGH
+
+normal_FAC1_check_mantissa_byte:
+
+	lda FAC1_mantissa+0
+	beq normal_FAC1_by_byte                 // branch if we can move the whole byte
 	
+	// FALLTROUGH
+
+normal_FAC1_check_mantissa_bit:
+
+	lda FAC1_mantissa+0                     // XXX optimize for 65CE02?
 	and #$80
-	bne normal_FAC1_end                // branch if already normalized
+	bne normal_FAC1_end                     // branch if already normalized
 	
 	// FALLTROUGH
 	
 normal_FAC1_by_bit:
 	
-	// First check the current exponent
+	// Decrement the exponent
 	
-	lda FAC1_exponent
-	cmp #$01
-	bne !+
-	
-	// Exponent is 1 - just mark the whole number as 0 and finish
-	
-	lda #$00
-	sta FAC1_exponent
-	rts
-!:
+	dec FAC1_exponent
+	beq normal_FAC1_end                     // exponent reached 0, nothing more to do
+
 	// Multiply the mantissa by 2
 
 	clc
@@ -49,7 +51,7 @@ normal_FAC1_by_bit:
 	rol FAC1_mantissa+1
 	rol FAC1_mantissa+0
 	
-	bcc normal_FAC1                     // branch if next iteration needed
+	bcc normal_FAC1_check_mantissa_bit      // branch if next iteration needed
 	rts
 	
 normal_FAC1_by_byte:
@@ -63,19 +65,19 @@ normal_FAC1_by_byte:
 	lda #$00
 !:
 	sta FAC1_exponent
-	beq normal_FAC1_end                 // branch if our float reached 0
+	beq normal_FAC1_end                      // branch if our float reached 0
 
 	// Now multiply the mantissa by 8 by moving the whole bytes
 	
-	lda FAC1_mantissa+2
-	sta FAC1_mantissa+3
 	lda FAC1_mantissa+1
-	sta FAC1_mantissa+2
-	lda FAC1_mantissa+0
+	sta FAC1_mantissa+0
+	lda FAC1_mantissa+2
 	sta FAC1_mantissa+1
+	lda FAC1_mantissa+3
+	sta FAC1_mantissa+2
 	
 	lda FACOV
-	sta FAC1_mantissa+0
+	sta FAC1_mantissa+3
 
 	lda #$00
 	sta FACOV
@@ -87,7 +89,7 @@ normal_FAC1_by_byte:
 	ora FAC1_mantissa+2
 	ora FAC1_mantissa+3
 	
-	bne normal_FAC1                     // not 0 - branch to next iteration
+	bne normal_FAC1_check_mantissa_byte     // not 0 - branch to next iteration
 
 	// Our mantissa reached 0 - mark exponent as 0 and finish