Add ISA definitions for symphony

Adds both a 'stock' ISA (copied from the game) and one modified for test
code

isa/README.md

@@ -16,7 +16,7 @@ Performs a jump-and-link function call, with label size up to 32 bits.
 
 _HARDWARE:_ Store PC+4 in the result register when in jump mode.
 
-## `li %a(register) %b:U32(immediate | label)`
+## `li %a(register) %b:U32(immediate | label)
 Loads an immediate up to 32 bits long
 
 _HARDWARE:_ No requirements

isa/test.isa

@@ -332,13 +332,12 @@ ret
 li %r(register), %a:S33(immediate | label)
 ; assert(0 - 0x80000000 <= %a, "%a is too low")
 ; assert(%a <= 0xffffffff, "%a is too high")
-%s = trailing_zeros(%a)
+%nota = ~%a
-%sa = %a >> %s
+when 0x0000 <= %a && %a <= 0xffff: 00110001 rrrr0000 %a[15:0]                                           ; First 64k
-%nota = %a ^ 0xffffffff
+when %nota <= 0xffff             : 00110001 rrrr0000 %nota[15:0]    00100011 rrrr0000 0000rrrr 00000000 ; Last 64k
-when %a <=u 0xffff   : 00110001 %r[3:0]0000 %a[15:0]                                                 ; First 64k
+when !(%a & 0x0f) && %a < 0xffff0: 00110001 rrrr0000 %a[19:4]       00100111 rrrrrrrr 00000000 00000100 ; 16-byte aligned
-when %nota <=u 0xffff: 00110001 %r[3:0]0000 %nota[15:0]    00100011 %r[3:0]0000 0000%r[3:0] 00000000 ; Last 64k
+when !(%a & 0x03) && %a < 0x3fffc: 00110001 rrrr0000 %a[17:2]       00100111 rrrrrrrr 00000000 00000010 ; 4-byte aligned
-when %sa <=u 0xffff  : 00110001 %r[3:0]0000 %sa[15:0]      00110111 %r[3:0]%r[3:0] 00000000 %s[7:0]  ; Aligned
+00110001 rrrr0000 %a[31:16]   00100111 rrrrrrrr 00000000 00010000   00110001 rrrrrrrr %a[15:0]          ; Any 32-bit
-00110001 %r[3:0]0000 %a[31:16]   00110111 %r[3:0]%r[3:0] 00000000 00010000   00110001 %r[3:0]%r[3:0] %a[15:0]          ; Any 32-bit
 # Load 32-bit immediate %a into %r
 
 qcall %a:U32(label)
@@ -346,16 +345,14 @@ qcall %a:U32(label)
 ; assert(%a <= 0xffffffff, "%a is too high")
 %r = 13
 %f = 15
-%s = trailing_zeros(%a)
+%nota = ~%a
-%sa = %a >> %s
+when 0x0000 <= %a && %a <= 0xffff: 01011000 rrrrffff %a[15:0]                                                                                  ; First 64k
-%nota = %a ^ 0xffffffff
+when %nota <= 0xffff             : 00110001 rrrr0000 %nota[15:0]    00100011 rrrr0000 0000rrrr 00000000    01001000 rrrrffff 0000rrrr 00000000 ; Last 64k
-when %a <=u 0xffff   : 01011000 %r[3:0]ffff %a[15:0]                                                                                              ; First 64k
+when !(%a & 0x0f) && %a < 0xffff0: 00110001 rrrr0000 %a[19:4]       00100111 rrrrrrrr 00000000 00000100    01001000 rrrrffff 0000rrrr 00000000 ; 16-byte aligned
-when %nota <=u 0xffff: 00110001 %r[3:0]0000 %nota[15:0]    00100011 %r[3:0]0000 0000%r[3:0] 00000000    01001000 %r[3:0]ffff 0000%r[3:0] 00000000 ; Last 64k
+when !(%a & 0x03) && %a < 0x3fffc: 00110001 rrrr0000 %a[17:2]       00100111 rrrrrrrr 00000000 00000010    01001000 rrrrffff 0000rrrr 00000000 ; 4-byte aligned
-when %sa <=u 0xffff  : 00110001 %r[3:0]0000 %sa[15:0]      00110111 %r[3:0]%r[3:0] 00000000 %s[7:0]     01001000 %r[3:0]ffff 0000%r[3:0] 00000000 ; Aligned
+00110001 rrrr0000 %a[31:16]   00100111 rrrrrrrr 00000000 00010000   00110001 rrrrrrrr %a[15:0]             01001000 rrrrffff 0000rrrr 00000000 ; Any 32-bit
-00110001 %r[3:0]0000 %a[31:16]   00110111 %r[3:0]%r[3:0] 00000000 00010000   00110001 %r[3:0]%r[3:0] %a[15:0]    01001000 %r[3:0]ffff 0000%r[3:0] 00000000 ; Any 32-bit
 # Jump and link to 32-bit label
 
 halt%c(condition)
 1000cccc 00001111 00000000 00000000
 # Halt on condition %c
-

src/imath.asm

@@ -4,53 +4,22 @@
 ; r2 - The second value
 ; Result:
 ; r1 - The lower 32 bits of the result
-; Clobbers: r2, r3, r4, r5, r6
+; Clobbers: r2, r3, r4, r5
 pub mul_low:
-	cmp r1, r2
+	mov r3, 0 ; result
-	jbe mul_low_noswap
+	mov r4, 31 ; loop counter
-	xor r1, r2, r1
+	
-	xor r2, r1, r2
+	mul_low_loop:
-	xor r1, r2, r1
+		asr r5, r2, 31
-	; Passthrough
+		and r5, r5, r1
-
+		lsl r5, r5, r4
-; Multiplies r1 and r2, returning the lower part of the result
+		add r3, r3, r5
-; This method assumes r1 is smaller than r2, which results in faster execution
+		lsl r2, r2, 1
-; Arguments:
+		sub r4, r4, 1
-; r1 - The first value
+		cmp r4, 0
-; r2 - The second value
+		jge mul_low_loop
-; Result:
+	mov r1, r3
-; r1 - The lower 32 bits of the result
-; Clobbers: r2, r3, r4, r5, r6
-pub mul_low_noswap:
-	mov r4, 0 ; r4 has the result
-	mov r6, mul_loop_end
-	add r3, r2, r2
 	
-	mul_loop:
-		and r5, r1, 14 ; Taking the first four bits, discarding the odd
-		lsl r5, r5, 1 ; 0000 - 0, 0001 - 2, 0010 - 4, 0011 - 8, etc.
-		sub r5, r6, r5
-		jmp r5
-		
-		add r4, r4, r3
-		add r4, r4, r3
-		add r4, r4, r3
-		add r4, r4, r3
-		add r4, r4, r3
-		add r4, r4, r3
-		add r4, r4, r3
-		mul_loop_end:
-			mov flags, r1
-			jne mul_skip_one
-			add r4, r4, r2
-			mul_skip_one:
-			lsl r2, r2, 4
-			lsl r3, r3, 4
-			lsr r1, r1, 4
-			cmp r1, zr
-			jne mul_loop
-
-	mov r1, r4
 	jmp r13
 
 ; Calculates the absolute value of the value provided in the r1 register

src/mem.asm

@@ -1,174 +0,0 @@
-; int compare(uint8_t* a, uint8_t* b, size_t count);
-;   Compares two memory segment of equal length lexicographically.
-;
-; Arguments:
-; - `r1`: A pointer to the first memory segment.
-; - `r2`: A pointer to the second memory segment.
-; - `r3`: The size of both memory segments.
-; Results:
-; - `r1`:
-;   -  `0` if both segments are equal.
-;   - `<0` if the first segment is less than the second segment.
-;   - `>0` if the first segment is greater than the second segment.
-;
-pub compare:
-    ; Exclusive end point of the first segment.
-    add r3, r3, r1
-    sub r3, r3, 4
-_compare__loop:
-    load_32 r4, [r1]
-    add r1, r1, 4
-    load_32 r5, [r2]
-    add r2, r2, 4
-    ; Comparing two sequences of 4 bytes lexicographically is equivalent to
-    ; comparing the corresponding big endian 32 bit words.
-    cmp r4, r5
-    jne _compare__break
-    ; Check if there are enough bytes left to continue with the vectorized loop.
-    cmp r1, r3
-    jbe _compare__loop
-    ; `r3 + 4 - r1 = <remaining byte count> = r3 - r1 mod 4`
-    sub flags, r3, r1
-    ; Check if one of the lowest 2 bits is non-zero
-    jbe _compare__rem
-    ; If not, we are done. Both segments are equal.
-    mov r1, 0
-    jmp r13
-_compare__break:
-    ; `flags` is the comparison result in the format of `cmp`. Convert it to the desired format.
-    ; 00 => 0x40000000 > 0
-    ; 01 => 0x00000000 = 0
-    ; 10 => 0xC0000000 < 0
-    xor r1, flags, 1
-    lsl r1, r1, 30
-    jmp r13
-_compare__rem:
-    ; Compute `S = 8*(4 - <remaining byte count>)` and
-    ; [r1] >> S, [r2] >> S
-    mov r3, 8
-    load_32 r4, [r1]
-    sub r3, r3, flags
-    load_32 r5, [r2]
-    lsl r3, r3, 3
-    lsr r4, r4, r3
-    lsr r5, r5, r3
-    ; Compare both values, now with garbage bytes removed.
-    cmp r4, r5
-    jmp _compare__break
-
-
-; void copy(void* src, void* dest, size_t count);
-;   Copies `count` bytes from `src` to `dest`. The two memory segments must not overlap.
-;
-; Arguments:
-; - `r1`: Pointer to the memory segment to be copied.
-; - `r2`: Pointer to the memory segment to be copied into.
-; - `r3`: Byte size of both the `src` and `dest` segments.
-;
-pub copy:
-    ; Exclusive end point of the source segment.
-    add r3, r1, r3
-    ; Last index from where we can safely copy 8 bytes per loop iteration.
-    sub r3, r3, 8
-    jmp _copy__loop_entry
-_copy__loop:
-    ; Copy 8 bytes from `src` to `dest`.
-    load_32 flags, [r1]
-    add r1, r1, 4
-    store_32 [r2], flags
-    add r2, r2, 4
-    load_32 flags, [r1]
-    add r1, r1, 4
-    store_32 [r2], flags
-    add r2, r2, 4
-_copy__loop_entry:
-    ; Check if we can process more data in the vectorized loop.
-    cmp r1, r3
-    jbe _copy__loop
-    ; The remaining amount of bytes `R` is `R = r3 + 8 - r1 = r3 - r1 mod 8`.
-    sub flags, r3, r1
-    ; Test if `R` is not a multiple of `4`, i.e. the lowest 2 bits are non-zero.
-    jbe _copy__rem
-    ; `R` is a multiple of `4`. Special case this.
-    ; Check if `R` is `0`, i.e. the third bit is also 0. In that case, we are already done.
-    ; There are no conditional indirect jumps, so we can't return immediately.
-    jge _copy__ret
-    ; `R = 4`. No need to update `r1` or `r2`, we don't need them anymore.
-    load_32 flags, [r1]
-    store_32 [r2], flags
-_copy__ret:
-    ; Return
-    jmp r13
-_copy__rem:
-    ; Optimize the remaining cases for code size.
-    ; End point of the source segment.
-    add r3, r3, 8
-    ; We already handled the case `R = 0` earlier,
-    ; so no bounds check needed for the first iteration.
-_copy__rem_loop:
-    ; Copy 1 byte.
-    load_8 flags, [r1]
-    add r1, r1, 1
-    store_8 [r2], flags
-    add r2, r2, 1
-    ; Check if we are still within the bounds.
-    cmp r1, r3
-    jb _copy__rem_loop
-    jmp r13
-    
-; void fill32(uint8_t* dest, size_t count, uint32_t value);
-;   Fills `count` bytes in `dest` with `value`. If `count` is not a multiple of 4, 
-;   the least significant bytes of `value` are cut off for the last entry.
-;
-; Arguments:
-; - `r1`: A pointer to the destination segment.
-; - `r2`: The size of the destination segment.
-; - `r3`: The 32 bit value that the segment is filled with.
-;
-pub fill32:
-    ; Exclusive end point of the destination segment.
-    add r2, r1, r2
-    ; Last index from where we can safely write 8 bytes per loop iteration.
-    sub r2, r2, 8
-    jmp _fill32__entry
-_fill32__loop:
-	; Set 8 bytes per loop iteraion.
-    store_32 [r1], r3
-    add r1, r1, 4
-    store_32 [r1], r3
-    add r1, r1, 4
-_fill32__entry:
-    ; Check if we can process more data in the vectorized loop.
-    cmp r1, r2
-    jbe _fill32__loop
-    ; The remaining amount of bytes `R` is `R = r2 + 8 - r1 = r2 - r1 mod 8`.
-    sub flags, r2, r1
-    ; Check if the third bit of the remainder is cleared.
-    jge _fill32__r4
-    ; Otherwise set 4 bytes.
-    store_32 [r1], r3
-    add r1, r1, 4
-_fill32__r4:
-    ; Check if the two least significant bits of the remainder are zero.
-    ja _fill32__ret
-    ; Handle the remaining bytes `R` individually, in reverse order.
-    add r2, r2, 4
-    ; `r1 + 4 - r2 = 4 - R`.
-    sub flags, r1, r2
-    ; Exclusive end point of the destination segment.
-    add r2, r2, 4
-    ; Shift out the least significant `8*(4 - R)` bits of the value.
-    lsl flags, flags, 3
-    lsr r3, r3, flags
-    jmp _fill32__loop2_entry
-_fill32__loop2:
-    sub r2, r2, 1
-    ; Write the least significant byte of the value...
-    store_8 [r2], r3
-    ; and then shift it out.
-    lsr r3, r3, 8
-_fill32__loop2_entry:
-    cmp r1, r2
-    jb _fill32__loop2
-_fill32__ret:
-    jmp r13

src/stdlib.asm

@@ -2,7 +2,6 @@ pub include bit
 pub include imath
 pub include array
 pub include console
-pub include mem
 
 ; Needs to be last!
 pub include LUTs

src/unsafe_mem.asm

@@ -1,77 +0,0 @@
-; int compare(uint8_t* a, uint8_t* b, size_t count);
-;   Compares two memory segment of equal length lexicographically.
-;   Temporarily modifies the byte at address `a + count`.
-;
-; Arguments:
-; - `r1`: A pointer to the first memory segment.
-; - `r2`: A pointer to the second memory segment.
-; - `r3`: The size of both memory segments.
-; Results:
-; - `r1`:
-;   -  `0` if both segments are equal.
-;   - `<0` if the first segment is less than the second segment.
-;   - `>0` if the first segment is greater than the second segment.
-pub compare:
-    cmp r1, r2
-    je _compare__is_eq
-    ; Exclusive end point of the second segment.
-    add r4, r3, r2
-    ; Exclusive end point of the first segment.
-    add r3, r3, r1
-    load_8 r6, [r3]
-    load_8 flags, [r4]
-    ; Check if the first bytes behind the sequences are equal.
-    cmp flags, r6
-    jne _compare__loop
-    ; Change the byte directly behind the first segment.
-    xor r4, r6, 1
-    ; This would be problematic if someone calls compare with a first segment
-    ; whose end point overlaps the program memory of this function.
-    store_8 [r3], r4
-_compare__loop:
-    load_32 r4, [r1]
-    add r1, r1, 4
-    load_32 r5, [r2]
-    add r2, r2, 4
-    ; Comparing two sequences of 4 bytes lexicographically is equivalent to
-    ; comparing the corresponding big endian 32 bit words.
-    cmp r4, r5
-    je _compare__loop
-    ; We overshot in the loop; decrement r1 again. (Only by 2, we backtrack the rest if necessary later)
-    sub r1, r1, 2
-    ; Restore the byte we changed.
-    store_8 [r3], r6
-    ; We encountered two different words. Figure out what byte they differ on.
-    xor r4, r4, r5
-    ; Store the flags for later, to figure out the return value.
-    mov r5, flags
-    ; Check if at least one of the two most significant bytes is not 0.
-    cmp r4, 0xffff
-    jbe _compare__low2
-    ; If it is, backtrack the remaining 2 indices.
-    ; Shift the most significant bytes to the least significant ones.
-    sub r1, r1, 2
-    lsr r4, r4, 16
-_compare__low2:
-    ; r1 now points to a non-zero 16 bit value.
-    ; If the 16 bit value at r1-2 is in-bounds, then it is 0.
-    ; Check if the most significant byte of the 16 bit value is 0.
-    cmp r4, 0xff
-    ja _compare__low1
-    ; If it is, our target is the least significant byte.
-    add r1, r1, 1
-_compare__low1:
-    ; Otherwise, the target is that non-zero byte.
-    ; Check if the target is out of bounds, i.e. the loop terminated through the "bounds check".
-    cmp r1, r3
-    jae _compare__is_eq
-    ; r5 is the comparison result in the format of `cmp`. Convert it to the desired format.
-    ; 00 => 0x40000000 > 0
-    ; 01 => 0x00000000 = 0
-    ; 10 => 0xC0000000 < 0
-    xor r1, r5, 1
-    lsl r1, r1, 30
-    jmp r13
-_compare__is_eq:
-    mov r1, 0
-    jmp r13