Mercurial > repos > blastem

changeset 155:94a65fb4e1c7

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Don't use the native call stack for M68K calls by default
author Mike Pavone <pavone@retrodev.com>
date Fri, 04 Jan 2013 23:21:07 -0800
parents 4791c0204410
children 3900cfde9dbb
files m68k_to_x86.c m68k_to_x86.h
diffstat 2 files changed, 88 insertions(+), 33 deletions(-) [+]
line wrap: on
line diff
--- a/m68k_to_x86.c	Fri Jan 04 22:51:01 2013 -0800
+++ b/m68k_to_x86.c	Fri Jan 04 23:21:07 2013 -0800
@@ -1398,7 +1398,9 @@
 	//TODO: Add cycles in the right place relative to pushing the return address on the stack
 	dst = cycles(dst, 10);
 	dst = mov_ir(dst, after, SCRATCH1, SZ_D);
-	dst = push_r(dst, SCRATCH1);
+	if (opts->flags & OPT_NATIVE_CALL_STACK) {
+		dst = push_r(dst, SCRATCH1);
+	}
 	dst = sub_ir(dst, 4, opts->aregs[7], SZ_D);
 	dst = mov_rr(dst, opts->aregs[7], SCRATCH2, SZ_D);
 	dst = call(dst, (char *)m68k_write_long_highfirst);
@@ -1406,11 +1408,15 @@
 	if (!dest_addr) {
 		opts->deferred = defer_address(opts->deferred, (inst->address+2) + disp, dst + 1);
 		//dummy address to be replaced later
-		dest_addr = dst + 5;
+		dest_addr = dst + 256;
 	}
-	dst = call(dst, (char *)dest_addr);
-	//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
-	dst = pop_r(dst, SCRATCH1);
+	if (opts->flags & OPT_NATIVE_CALL_STACK) {
+		dst = call(dst, (char *)dest_addr);
+		//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
+		dst = pop_r(dst, SCRATCH1);
+	} else {
+		dst = jmp(dst, (char *)dest_addr);
+	}
 	return dst;
 }
 
@@ -1725,7 +1731,9 @@
 	case MODE_AREG_INDIRECT:
 		dst = cycles(dst, BUS*2);
 		dst = mov_ir(dst, inst->address + 2, SCRATCH1, SZ_D);
-		dst = push_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			dst = push_r(dst, SCRATCH1);
+		}
 		dst = sub_ir(dst, 4, opts->aregs[7], SZ_D);
 		dst = mov_rr(dst, opts->aregs[7], SCRATCH2, SZ_D);
 		dst = call(dst, (char *)m68k_write_long_highfirst);
@@ -1735,14 +1743,20 @@
 			dst = mov_rdisp8r(dst, CONTEXT, offsetof(m68k_context, aregs) + 4 * inst->src.params.regs.pri, SCRATCH1, SZ_D);
 		}
 		dst = call(dst, (uint8_t *)m68k_native_addr);
-		dst = call_r(dst, SCRATCH1);
-		//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
-		dst = pop_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			dst = call_r(dst, SCRATCH1);
+			//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
+			dst = pop_r(dst, SCRATCH1);
+		} else {
+			dst = jmp_r(dst, SCRATCH1);
+		}
 		break;
 	case MODE_AREG_INDEX_DISP8:
 		dst = cycles(dst, BUS*3);//TODO: CHeck that this is correct
 		dst = mov_ir(dst, inst->address + 4, SCRATCH1, SZ_D);
-		dst = push_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			dst = push_r(dst, SCRATCH1);
+		}
 		dst = sub_ir(dst, 4, opts->aregs[7], SZ_D);
 		dst = mov_rr(dst, opts->aregs[7], SCRATCH2, SZ_D);
 		dst = call(dst, (char *)m68k_write_long_highfirst);
@@ -1786,15 +1800,21 @@
 			dst = add_ir(dst, inst->src.params.regs.displacement, SCRATCH1, SZ_D);
 		}
 		dst = call(dst, (uint8_t *)m68k_native_addr);
-		dst = call_r(dst, SCRATCH1);
-		//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
-		dst = pop_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			dst = call_r(dst, SCRATCH1);
+			//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
+			dst = pop_r(dst, SCRATCH1);
+		} else {
+			dst = jmp_r(dst, SCRATCH1);
+		}
 		break;
 	case MODE_PC_DISPLACE:
 		//TODO: Add cycles in the right place relative to pushing the return address on the stack
 		dst = cycles(dst, 10);
 		dst = mov_ir(dst, inst->address + 4, SCRATCH1, SZ_D);
-		dst = push_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			dst = push_r(dst, SCRATCH1);
+		}
 		dst = sub_ir(dst, 4, opts->aregs[7], SZ_D);
 		dst = mov_rr(dst, opts->aregs[7], SCRATCH2, SZ_D);
 		dst = call(dst, (char *)m68k_write_long_highfirst);
@@ -1804,21 +1824,33 @@
 			if (!dest_addr) {
 				opts->deferred = defer_address(opts->deferred, m68k_addr, dst + 1);
 				//dummy address to be replaced later, make sure it generates a 4-byte displacement
-				dest_addr = dst + 5;
+				dest_addr = dst + 256;
 			}
-			dst = call(dst, (char *)dest_addr);
+			if (opts->flags & OPT_NATIVE_CALL_STACK) {
+				dst = call(dst, (char *)dest_addr);
+			} else {
+				dst = jmp(dst, dest_addr);
+			}
 		} else {
 			dst = mov_ir(dst, m68k_addr, SCRATCH1, SZ_D);
 			dst = call(dst, (uint8_t *)m68k_native_addr);
-			dst = call_r(dst, SCRATCH1);
+			if (opts->flags & OPT_NATIVE_CALL_STACK) {
+				dst = call_r(dst, SCRATCH1);
+			} else {
+				dst = jmp_r(dst, SCRATCH1);
+			}
 		}
-		//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
-		dst = pop_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
+			dst = pop_r(dst, SCRATCH1);
+		}
 		break;
 	case MODE_PC_INDEX_DISP8:
 		dst = cycles(dst, BUS*3);//TODO: CHeck that this is correct
 		dst = mov_ir(dst, inst->address + 4, SCRATCH1, SZ_D);
-		dst = push_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			dst = push_r(dst, SCRATCH1);
+		}
 		dst = sub_ir(dst, 4, opts->aregs[7], SZ_D);
 		dst = mov_rr(dst, opts->aregs[7], SCRATCH2, SZ_D);
 		dst = call(dst, (char *)m68k_write_long_highfirst);
@@ -1858,16 +1890,22 @@
 			dst = add_ir(dst, inst->src.params.regs.displacement, SCRATCH1, SZ_D);
 		}
 		dst = call(dst, (uint8_t *)m68k_native_addr);
-		dst = call_r(dst, SCRATCH1);
-		//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
-		dst = pop_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			dst = call_r(dst, SCRATCH1);
+			//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
+			dst = pop_r(dst, SCRATCH1);
+		} else {
+			dst = jmp_r(dst, SCRATCH1);
+		}
 		break;
 	case MODE_ABSOLUTE:
 	case MODE_ABSOLUTE_SHORT:
 		//TODO: Add cycles in the right place relative to pushing the return address on the stack
 		dst = cycles(dst, inst->src.addr_mode == MODE_ABSOLUTE ? 12 : 10);
 		dst = mov_ir(dst, inst->address + (inst->src.addr_mode == MODE_ABSOLUTE ? 6 : 4), SCRATCH1, SZ_D);
-		dst = push_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			dst = push_r(dst, SCRATCH1);
+		}
 		dst = sub_ir(dst, 4, opts->aregs[7], SZ_D);
 		dst = mov_rr(dst, opts->aregs[7], SCRATCH2, SZ_D);
 		dst = call(dst, (char *)m68k_write_long_highfirst);
@@ -1877,16 +1915,26 @@
 			if (!dest_addr) {
 				opts->deferred = defer_address(opts->deferred, m68k_addr, dst + 1);
 				//dummy address to be replaced later, make sure it generates a 4-byte displacement
-				dest_addr = dst + 5;
+				dest_addr = dst + 256;
 			}
-			dst = call(dst, (char *)dest_addr);
+			if (opts->flags & OPT_NATIVE_CALL_STACK) {
+				dst = call(dst, (char *)dest_addr);
+			} else {
+				dst = jmp(dst, dest_addr);
+			}
 		} else {
 			dst = mov_ir(dst, m68k_addr, SCRATCH1, SZ_D);
 			dst = call(dst, (uint8_t *)m68k_native_addr);
-			dst = call_r(dst, SCRATCH1);
+			if (opts->flags & OPT_NATIVE_CALL_STACK) {
+				dst = call_r(dst, SCRATCH1);
+			} else {
+				dst = jmp_r(dst, SCRATCH1);
+			}
 		}
-		//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
-		dst = pop_r(dst, SCRATCH1);
+		if (opts->flags & OPT_NATIVE_CALL_STACK) {
+			//would add_ir(dst, 8, RSP, SZ_Q) be faster here?
+			dst = pop_r(dst, SCRATCH1);
+		}
 		break;
 	default:
 		m68k_disasm(inst, disasm_buf);
@@ -1902,10 +1950,15 @@
 	dst = mov_rr(dst, opts->aregs[7], SCRATCH1, SZ_D);
 	dst = add_ir(dst, 4, opts->aregs[7], SZ_D);
 	dst = call(dst, (char *)m68k_read_long_scratch1);
-	dst = cmp_rdisp8r(dst, RSP, 8, SCRATCH1, SZ_D);
-	dst = jcc(dst, CC_NZ, dst+3);
-	dst = retn(dst);
-	dst = jmp(dst, (char *)m68k_modified_ret_addr);
+	if (opts->flags & OPT_NATIVE_CALL_STACK) {
+		dst = cmp_rdisp8r(dst, RSP, 8, SCRATCH1, SZ_D);
+		dst = jcc(dst, CC_NZ, dst+3);
+		dst = retn(dst);
+		dst = jmp(dst, (char *)m68k_modified_ret_addr);
+	} else {
+		dst = call(dst, (uint8_t *)m68k_native_addr);
+		dst = jmp_r(dst, SCRATCH1);
+	}
 	return dst;
 }
 
--- a/m68k_to_x86.h	Fri Jan 04 22:51:01 2013 -0800
+++ b/m68k_to_x86.h	Fri Jan 04 23:21:07 2013 -0800
@@ -6,6 +6,8 @@
 #define NATIVE_CHUNK_SIZE ((16 * 1024 * 1024 / NATIVE_MAP_CHUNKS)/2)
 #define INVALID_OFFSET 0xFFFFFFFF
 
+#define OPT_NATIVE_CALL_STACK 0x1
+ 
 typedef struct {
 	uint8_t  *base;
 	int32_t *offsets;