Macintosh-Tools/gb6
1
0
Fork 0
mirror of https://github.com/mlaux/gb6.git synced 2026-04-26 17:18:02 +00:00
Code Issues Projects Releases Wiki Activity

add extra cycles for taken branch

Browse Source
This commit is contained in:
Matthew Laux
2026-02-16 17:05:48 -06:00
parent 28f4ff2d02
commit 8e30a2bff5
1 changed files with 51 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files
compiler/branches.c
+51 -31
View File
@@ -150,13 +150,15 @@ void compile_jr_cond(
// Tiny loops (disp >= -3): skip interrupt check, just branch
if (disp >= -3) {
// bxx.w displacement needs adjustment for where we emit it
int16_t cond_disp = (int16_t) target_m68k - (int16_t) (block->length + 2);
// Skip if NOT taken (skip: addq(2) + bra.w(4) = 6, +2 = 8)
if (branch_if_set) {
emit_bne_w(block, cond_disp);
emit_beq_w(block, 8);
} else {
emit_beq_w(block, cond_disp);
emit_bne_w(block, 8);
}
emit_add_cycles(block, 4); // extra cycles for taken branch
int16_t bra_disp = (int16_t) target_m68k - (int16_t) (block->length + 2);
emit_bra_w(block, bra_disp);
return;
}
@@ -185,10 +187,11 @@ void compile_jr_cond(
emit_beq_w(block, 6);
}
// bra.w to .fall_through (cmpi.l(6) + bcs.w(4) + moveq(2) + move.w(4) + patchable_exit(16) = 32, plus 2 for PC = 34)
emit_bra_w(block, 34);
// bra.w to .fall_through (addq(2) + cmpi.l(6) + bcs.w(4) + moveq(2) + move.w(4) + patchable_exit(16) = 34, plus 2 for PC = 36)
emit_bra_w(block, 36);
// .check_cycles:
emit_add_cycles(block, 4); // extra cycles for taken branch
emit_cmpi_l_imm_dn(block, cycles_per_exit, REG_68K_D_CYCLE_COUNT);
// bcs.w to native loop target (cycles < cycles_per_exit)
@@ -210,12 +213,13 @@ void compile_jr_cond(
if (branch_if_set) {
// Skip exit if flag is clear (btst Z=1 when bit=0)
emit_beq_w(block, 24); // skip: moveq(2) + move.w(4) + patchable_exit(16) = 22, plus 2 = 24
emit_beq_w(block, 26); // skip: addq(2) + moveq(2) + move.w(4) + patchable_exit(16) = 24, plus 2 = 26
} else {
// Skip exit if flag is set (btst Z=0 when bit=1)
emit_bne_w(block, 24);
emit_bne_w(block, 26);
}
emit_add_cycles(block, 4); // extra cycles for taken branch
emit_moveq_dn(block, REG_68K_D_NEXT_PC, 0);
emit_move_w_dn(block, REG_68K_D_NEXT_PC, target_gb_pc);
emit_patchable_exit(block);
@@ -241,12 +245,13 @@ void compile_jp_cond(
// If condition NOT met, skip the exit sequence
if (branch_if_set) {
// Skip exit if flag is clear (btst Z=1 when bit=0)
emit_beq_w(block, 24); // skip: moveq(2) + move.w(4) + patchable_exit(16) = 22, plus 2 = 24
emit_beq_w(block, 26); // skip: addq(2) + moveq(2) + move.w(4) + patchable_exit(16) = 24, plus 2 = 26
} else {
// Skip exit if flag is set (btst Z=0 when bit=1)
emit_bne_w(block, 24);
emit_bne_w(block, 26);
}
emit_add_cycles(block, 4); // extra cycles for taken branch
emit_moveq_dn(block, REG_68K_D_NEXT_PC, 0);
emit_move_w_dn(block, REG_68K_D_NEXT_PC, target);
emit_patchable_exit(block);
@@ -295,14 +300,16 @@ void compile_call_cond(
emit_btst_imm_dn(block, flag_bit, REG_68K_D_FLAGS);
// If condition NOT met, skip the call sequence
// call sequence: subq(2) + subi(6) + move.w(4) + move.b(2) + rol(2) + move.b(4) +
// moveq(2) + move.w(4) + patchable_exit(16) = 42, +2 = 44
// call sequence: addi(6) + subq(2) + subi(6) + move.w(4) + move.b(2) + rol(2) + move.b(4) +
// moveq(2) + move.w(4) + patchable_exit(16) = 48, +2 = 50
if (branch_if_set) {
emit_beq_w(block, 44);
emit_beq_w(block, 50);
} else {
emit_bne_w(block, 44);
emit_bne_w(block, 50);
}
emit_add_cycles(block, 12); // extra cycles for taken branch
// Push return address
emit_subq_w_an(block, REG_68K_A_SP, 2);
emit_subi_w_disp_an(block, 2, JIT_CTX_GB_SP, REG_68K_A_CTX);
@@ -338,14 +345,16 @@ void compile_ret_cond(struct code_block *block, uint8_t flag_bit, int branch_if_
emit_btst_imm_dn(block, flag_bit, REG_68K_D_FLAGS);
// If condition NOT met, skip the return sequence
// ret sequence: moveq(2) + move.b(4) + rol(2) + move.b(2) + addq(2) + addi(6) +
// dispatch_jump(6) = 24, +2 = 26
// ret sequence: addi(6) + moveq(2) + move.b(4) + rol(2) + move.b(2) + addq(2) + addi(6) +
// dispatch_jump(6) = 30, +2 = 32
if (branch_if_set) {
emit_beq_w(block, 26);
emit_beq_w(block, 32);
} else {
emit_bne_w(block, 26);
emit_bne_w(block, 32);
}
emit_add_cycles(block, 12); // extra cycles for taken branch
// Pop return address and dispatch
emit_moveq_dn(block, REG_68K_D_NEXT_PC, 0);
emit_move_b_disp_an_dn(block, 1, REG_68K_A_SP, REG_68K_D_NEXT_PC);
@@ -406,8 +415,11 @@ int compile_jr_cond_fused(
// Tiny loops: skip cycle check
if (disp >= -3) {
// Skip if NOT taken (skip: addq(2) + bra.w(4) = 6, +2 = 8)
emit_bcc_opcode_w(block, invert_cond(cond), 8);
emit_add_cycles(block, 4); // extra cycles for taken branch
m68k_disp = (int16_t) target_m68k - (int16_t) (block->length + 2);
emit_bcc_opcode_w(block, cond, m68k_disp);
emit_bra_w(block, m68k_disp);
return 0;
}
@@ -416,6 +428,7 @@ int compile_jr_cond_fused(
// bcc.w .check_cycles ; if condition met
// bra.w .fall_through ; condition not met
// .check_cycles:
// addq.l #4, d2 ; extra cycles for taken branch
// cmpi.l #cycles_per_exit, d2
// bcs.w loop_target ; cycles < cycles_per_exit
// <exit via patchable_exit>
@@ -424,10 +437,11 @@ int compile_jr_cond_fused(
// Branch to check_cycles if condition met
emit_bcc_opcode_w(block, cond, 6);
// bra.w to .fall_through (cmpi.l(6) + bcs.w(4) + exit(22) = 32, plus 2 = 34)
emit_bra_w(block, 34);
// bra.w to .fall_through (addq(2) + cmpi.l(6) + bcs.w(4) + exit(22) = 34, plus 2 = 36)
emit_bra_w(block, 36);
// .check_cycles:
emit_add_cycles(block, 4); // extra cycles for taken branch
emit_cmpi_l_imm_dn(block, cycles_per_exit, REG_68K_D_CYCLE_COUNT);
// bcs.w to native loop target (cycles < cycles_per_exit)
@@ -445,9 +459,10 @@ int compile_jr_cond_fused(
// Forward/external jump - conditionally exit via patchable exit
target_gb_pc = src_address + target_gb_offset;
// Skip exit if condition NOT met (skip: moveq(2) + move.w(4) + patchable_exit(16) = 22, +2 = 24)
emit_bcc_opcode_w(block, invert_cond(cond), 24);
// Skip exit if condition NOT met (skip: addq(2) + moveq(2) + move.w(4) + patchable_exit(16) = 24, +2 = 26)
emit_bcc_opcode_w(block, invert_cond(cond), 26);
emit_add_cycles(block, 4); // extra cycles for taken branch
emit_moveq_dn(block, REG_68K_D_NEXT_PC, 0);
emit_move_w_dn(block, REG_68K_D_NEXT_PC, target_gb_pc);
emit_patchable_exit(block);
@@ -465,9 +480,10 @@ void compile_jp_cond_fused(
uint16_t target = READ_BYTE(*src_ptr) | (READ_BYTE(*src_ptr + 1) << 8);
*src_ptr += 2;
// Skip exit if condition NOT met (skip: moveq(2) + move.w(4) + patchable_exit(16) = 22, +2 = 24)
emit_bcc_opcode_w(block, invert_cond(cond), 24);
// Skip exit if condition NOT met (skip: addq(2) + moveq(2) + move.w(4) + patchable_exit(16) = 24, +2 = 26)
emit_bcc_opcode_w(block, invert_cond(cond), 26);
emit_add_cycles(block, 4); // extra cycles for taken branch
emit_moveq_dn(block, REG_68K_D_NEXT_PC, 0);
emit_move_w_dn(block, REG_68K_D_NEXT_PC, target);
emit_patchable_exit(block);
@@ -477,9 +493,11 @@ void compile_jp_cond_fused(
void compile_ret_cond_fused(struct code_block *block, int cond)
{
// Skip return if condition NOT met
// ret sequence: moveq(2) + move.b(4) + rol(2) + move.b(2) + addq(2) + addi(6) +
// dispatch_jump(6) = 24, +2 = 26
emit_bcc_opcode_w(block, invert_cond(cond), 26);
// ret sequence: addi(6) + moveq(2) + move.b(4) + rol(2) + move.b(2) + addq(2) + addi(6) +
// dispatch_jump(6) = 30, +2 = 32
emit_bcc_opcode_w(block, invert_cond(cond), 32);
emit_add_cycles(block, 12); // extra cycles for taken branch
// Pop return address and dispatch
emit_moveq_dn(block, REG_68K_D_NEXT_PC, 0);
@@ -504,9 +522,11 @@ void compile_call_cond_fused(
*src_ptr += 2;
// Skip call if condition NOT met
// call sequence: subq(2) + subi(6) + move.w(4) + move.b(2) + rol(2) + move.b(4) +
// moveq(2) + move.w(4) + patchable_exit(16) = 42, +2 = 44
emit_bcc_opcode_w(block, invert_cond(cond), 44);
// call sequence: addi(6) + subq(2) + subi(6) + move.w(4) + move.b(2) + rol(2) + move.b(4) +
// moveq(2) + move.w(4) + patchable_exit(16) = 48, +2 = 50
emit_bcc_opcode_w(block, invert_cond(cond), 50);
emit_add_cycles(block, 12); // extra cycles for taken branch
// Push return address
emit_subq_w_an(block, REG_68K_A_SP, 2);