drc: rework block tracking and lookup
[picodrive.git] / cpu / drc / emit_arm.c
index 8b6af69..8b6c14d 100644 (file)
@@ -1,8 +1,10 @@
-// Basic macros to emit ARM instructions and some utils
-
-// (c) Copyright 2008-2009, Grazvydas "notaz" Ignotas
-// Free for non-commercial use.
-
+/*
+ * Basic macros to emit ARM instructions and some utils
+ * Copyright (C) 2008,2009,2010 notaz
+ *
+ * This work is licensed under the terms of MAME license.
+ * See COPYING file in the top-level directory.
+ */
 #define CONTEXT_REG 11
 
 // XXX: tcache_ptr type for SVP and SH2 compilers differs..
@@ -23,6 +25,7 @@
 #define A_R9M  (1 << 9)
 #define A_R10M (1 << 10)
 #define A_R11M (1 << 11)
+#define A_R12M (1 << 12)
 #define A_R14M (1 << 14)
 #define A_R15M (1 << 15)
 
 #define EOP_C_AM2_IMM(cond,u,b,l,rn,rd,offset_12) \
        EMIT(((cond)<<28) | 0x05000000 | ((u)<<23) | ((b)<<22) | ((l)<<20) | ((rn)<<16) | ((rd)<<12) | (offset_12))
 
+#define EOP_C_AM2_REG(cond,u,b,l,rn,rd,shift_imm,shift_op,rm) \
+       EMIT(((cond)<<28) | 0x07000000 | ((u)<<23) | ((b)<<22) | ((l)<<20) | ((rn)<<16) | ((rd)<<12) | \
+               ((shift_imm)<<7) | ((shift_op)<<5) | (rm))
+
 /* addressing mode 3 */
 #define EOP_C_AM3(cond,u,r,l,rn,rd,s,h,immed_reg) \
        EMIT(((cond)<<28) | 0x01000090 | ((u)<<23) | ((r)<<22) | ((l)<<20) | ((rn)<<16) | ((rd)<<12) | \
 #define EOP_C_AM3_REG(cond,u,l,rn,rd,s,h,rm)       EOP_C_AM3(cond,u,0,l,rn,rd,s,h,rm)
 
 /* ldr and str */
+#define EOP_LDR_IMM2(cond,rd,rn,offset_12)  EOP_C_AM2_IMM(cond,1,0,1,rn,rd,offset_12)
+#define EOP_LDRB_IMM2(cond,rd,rn,offset_12) EOP_C_AM2_IMM(cond,1,1,1,rn,rd,offset_12)
+
 #define EOP_LDR_IMM(   rd,rn,offset_12) EOP_C_AM2_IMM(A_COND_AL,1,0,1,rn,rd,offset_12)
 #define EOP_LDR_NEGIMM(rd,rn,offset_12) EOP_C_AM2_IMM(A_COND_AL,0,0,1,rn,rd,offset_12)
 #define EOP_LDR_SIMPLE(rd,rn)           EOP_C_AM2_IMM(A_COND_AL,1,0,1,rn,rd,0)
 #define EOP_STR_IMM(   rd,rn,offset_12) EOP_C_AM2_IMM(A_COND_AL,1,0,0,rn,rd,offset_12)
 #define EOP_STR_SIMPLE(rd,rn)           EOP_C_AM2_IMM(A_COND_AL,1,0,0,rn,rd,0)
 
+#define EOP_LDR_REG_LSL(cond,rd,rn,rm,shift_imm) EOP_C_AM2_REG(cond,1,0,1,rn,rd,shift_imm,A_AM1_LSL,rm)
+
+#define EOP_LDRH_IMM2(cond,rd,rn,offset_8)  EOP_C_AM3_IMM(cond,1,1,rn,rd,0,1,offset_8)
+
 #define EOP_LDRH_IMM(   rd,rn,offset_8)  EOP_C_AM3_IMM(A_COND_AL,1,1,rn,rd,0,1,offset_8)
 #define EOP_LDRH_SIMPLE(rd,rn)           EOP_C_AM3_IMM(A_COND_AL,1,1,rn,rd,0,1,0)
 #define EOP_LDRH_REG(   rd,rn,rm)        EOP_C_AM3_REG(A_COND_AL,1,1,rn,rd,0,1,rm)
 #define EOP_C_BX(cond,rm) \
        EMIT(((cond)<<28) | 0x012fff10 | (rm))
 
-#define EOP_BX(rm) EOP_C_BX(A_COND_AL,rm)
-
 #define EOP_C_B_PTR(ptr,cond,l,signed_immed_24) \
        EMIT_PTR(ptr, ((cond)<<28) | 0x0a000000 | ((l)<<24) | (signed_immed_24))
 
 #define EOP_MSR_REG(rm)       EOP_C_MSR_REG(A_COND_AL,rm)
 
 
+// XXX: AND, RSB, *C, will break if 1 insn is not enough
 static void emith_op_imm2(int cond, int s, int op, int rd, int rn, unsigned int imm)
 {
        int ror2;
        u32 v;
 
-       if (op == A_OP_MOV) {
+       switch (op) {
+       case A_OP_MOV:
                rn = 0;
-               if (~imm < 0x100) {
+               if (~imm < 0x10000) {
                        imm = ~imm;
                        op = A_OP_MVN;
                }
-       } else if (imm == 0)
-               return;
+               break;
+
+       case A_OP_EOR:
+       case A_OP_SUB:
+       case A_OP_ADD:
+       case A_OP_ORR:
+       case A_OP_BIC:
+               if (s == 0 && imm == 0)
+                       return;
+               break;
+       }
 
-       for (v = imm, ror2 = 0; v != 0 || op == A_OP_MOV; v >>= 8, ror2 -= 8/2) {
+       for (v = imm, ror2 = 0; ; ror2 -= 8/2) {
                /* shift down to get 'best' rot2 */
                for (; v && !(v & 3); v >>= 2)
                        ror2--;
 
                EOP_C_DOP_IMM(cond, op, s, rn, rd, ror2 & 0x0f, v & 0xff);
 
-               if (op == A_OP_MOV) {
+               v >>= 8;
+               if (v == 0)
+                       break;
+               if (op == A_OP_MOV)
                        op = A_OP_ORR;
-                       rn = rd;
-               }
+               if (op == A_OP_MVN)
+                       op = A_OP_BIC;
+               rn = rd;
        }
 }
 
@@ -308,8 +335,8 @@ static int emith_xbranch(int cond, void *target, int is_call)
        tcache_ptr += sizeof(u32)
 
 #define JMP_EMIT(cond, ptr) { \
-       int val = (u32 *)tcache_ptr - (u32 *)(ptr) - 2; \
-       EOP_C_B_PTR(ptr, cond, 0, val & 0xffffff); \
+       u32 val_ = (u32 *)tcache_ptr - (u32 *)(ptr) - 2; \
+       EOP_C_B_PTR(ptr, cond, 0, val_ & 0xffffff); \
 }
 
 #define EMITH_JMP_START(cond) { \
@@ -321,11 +348,14 @@ static int emith_xbranch(int cond, void *target, int is_call)
 }
 
 // fake "simple" or "short" jump - using cond insns instead
-#define EMITH_SJMP_START(cond) \
+#define EMITH_NOTHING1(cond) \
        (void)(cond)
 
-#define EMITH_SJMP_END(cond) \
-       (void)(cond)
+#define EMITH_SJMP_START(cond) EMITH_NOTHING1(cond)
+#define EMITH_SJMP_END(cond)   EMITH_NOTHING1(cond)
+#define EMITH_SJMP3_START(cond)        EMITH_NOTHING1(cond)
+#define EMITH_SJMP3_MID(cond)  EMITH_NOTHING1(cond)
+#define EMITH_SJMP3_END()
 
 #define emith_move_r_r(d, s) \
        EOP_MOV_REG_SIMPLE(d, s)
@@ -402,6 +432,9 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_add_r_imm(r, imm) \
        emith_op_imm(A_COND_AL, 0, A_OP_ADD, r, imm)
 
+#define emith_adc_r_imm(r, imm) \
+       emith_op_imm(A_COND_AL, 0, A_OP_ADC, r, imm)
+
 #define emith_sub_r_imm(r, imm) \
        emith_op_imm(A_COND_AL, 0, A_OP_SUB, r, imm)
 
@@ -461,6 +494,12 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_and_r_r_imm(d, s, imm) \
        emith_op_imm2(A_COND_AL, 0, A_OP_AND, d, s, imm)
 
+#define emith_add_r_r_imm(d, s, imm) \
+       emith_op_imm2(A_COND_AL, 0, A_OP_ADD, d, s, imm)
+
+#define emith_sub_r_r_imm(d, s, imm) \
+       emith_op_imm2(A_COND_AL, 0, A_OP_SUB, d, s, imm)
+
 #define emith_neg_r_r(d, s) \
        EOP_RSB_IMM(d, s, 0, 0)
 
@@ -473,8 +512,11 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_asr(d, s, cnt) \
        EOP_MOV_REG(A_COND_AL,0,d,s,A_AM1_ASR,cnt)
 
+#define emith_ror_c(cond, d, s, cnt) \
+       EOP_MOV_REG(cond,0,d,s,A_AM1_ROR,cnt)
+
 #define emith_ror(d, s, cnt) \
-       EOP_MOV_REG(A_COND_AL,0,d,s,A_AM1_ROR,cnt)
+       emith_ror_c(A_COND_AL, d, s, cnt)
 
 #define emith_rol(d, s, cnt) \
        EOP_MOV_REG(A_COND_AL,0,d,s,A_AM1_ROR,32-(cnt)); \
@@ -524,8 +566,26 @@ static int emith_xbranch(int cond, void *target, int is_call)
        EOP_C_SMLAL(A_COND_AL,0,dhi,dlo,s1,s2)
 
 // misc
+#define emith_read_r_r_offs_c(cond, r, rs, offs) \
+       EOP_LDR_IMM2(cond, r, rs, offs)
+
+#define emith_read8_r_r_offs_c(cond, r, rs, offs) \
+       EOP_LDRB_IMM2(cond, r, rs, offs)
+
+#define emith_read16_r_r_offs_c(cond, r, rs, offs) \
+       EOP_LDRH_IMM2(cond, r, rs, offs)
+
+#define emith_read_r_r_offs(r, rs, offs) \
+       emith_read_r_r_offs_c(A_COND_AL, r, rs, offs)
+
+#define emith_read8_r_r_offs(r, rs, offs) \
+       emith_read8_r_r_offs_c(A_COND_AL, r, rs, offs)
+
+#define emith_read16_r_r_offs(r, rs, offs) \
+       emith_read16_r_r_offs_c(A_COND_AL, r, rs, offs)
+
 #define emith_ctx_read(r, offs) \
-       EOP_LDR_IMM(r, CONTEXT_REG, offs)
+       emith_read_r_r_offs(r, CONTEXT_REG, offs)
 
 #define emith_ctx_write(r, offs) \
        EOP_STR_IMM(r, CONTEXT_REG, offs)
@@ -573,9 +633,6 @@ static int emith_xbranch(int cond, void *target, int is_call)
        EOP_MOV_REG_ASR(d,d,32 - (bits)); \
 }
 
-#define host_arg2reg(rd, arg) \
-       rd = arg
-
 // upto 4 args
 #define emith_pass_arg_r(arg, reg) \
        EOP_MOV_REG_SIMPLE(arg, reg)
@@ -583,36 +640,92 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_pass_arg_imm(arg, imm) \
        emith_move_r_imm(arg, imm)
 
-#define emith_call_cond(cond, target) \
-       emith_xbranch(cond, target, 1)
+#define emith_jump(target) \
+       emith_jump_cond(A_COND_AL, target)
+
+#define emith_jump_patchable(target) \
+       emith_jump(target)
 
 #define emith_jump_cond(cond, target) \
        emith_xbranch(cond, target, 0)
 
-#define emith_call(target) \
-       emith_call_cond(A_COND_AL, target)
-
-#define emith_jump(target) \
-       emith_jump_cond(A_COND_AL, target)
-
-#define emith_jump_patchable(cond) \
-       emith_jump_cond(cond, 0)
+#define emith_jump_cond_patchable(cond, target) \
+       emith_jump_cond(cond, target)
 
 #define emith_jump_patch(ptr, target) do { \
        u32 *ptr_ = ptr; \
-       u32 val = (u32 *)(target) - (u32 *)ptr_ - 2; \
-       *ptr_ = (*ptr_ & 0xff000000) | (val & 0x00ffffff); \
+       u32 val_ = (u32 *)(target) - ptr_ - 2; \
+       *ptr_ = (*ptr_ & 0xff000000) | (val_ & 0x00ffffff); \
 } while (0)
 
+#define emith_jump_at(ptr, target) { \
+       u32 val_ = (u32 *)(target) - (u32 *)(ptr) - 2; \
+       EOP_C_B_PTR(ptr, A_COND_AL, 0, val_ & 0xffffff); \
+}
+
+#define emith_jump_reg_c(cond, r) \
+       EOP_C_BX(cond, r)
+
 #define emith_jump_reg(r) \
-       EOP_BX(r)
+       emith_jump_reg_c(A_COND_AL, r)
+
+#define emith_jump_ctx_c(cond, offs) \
+       EOP_LDR_IMM2(cond,15,CONTEXT_REG,offs)
+
+#define emith_jump_ctx(offs) \
+       emith_jump_ctx_c(A_COND_AL, offs)
+
+#define emith_call_cond(cond, target) \
+       emith_xbranch(cond, target, 1)
+
+#define emith_call(target) \
+       emith_call_cond(A_COND_AL, target)
+
+#define emith_call_ctx(offs) { \
+       emith_move_r_r(14, 15); \
+       emith_jump_ctx(offs); \
+}
+
+#define emith_ret_c(cond) \
+       emith_jump_reg_c(cond, 14)
+
+#define emith_ret() \
+       emith_ret_c(A_COND_AL)
+
+#define emith_ret_to_ctx(offs) \
+       emith_ctx_write(14, offs)
+
+#define emith_push_ret() \
+       EOP_STMFD_SP(A_R14M)
+
+#define emith_pop_and_ret() \
+       EOP_LDMFD_SP(A_R15M)
+
+#define host_instructions_updated(base, end) \
+       cache_flush_d_inval_i(base, end)
+
+#define host_arg2reg(rd, arg) \
+       rd = arg
 
 /* SH2 drc specific */
+/* pushes r12 for eabi alignment */
 #define emith_sh2_drc_entry() \
-       EOP_STMFD_SP(A_R4M|A_R5M|A_R6M|A_R7M|A_R8M|A_R9M|A_R10M|A_R11M|A_R14M)
+       EOP_STMFD_SP(A_R4M|A_R5M|A_R6M|A_R7M|A_R8M|A_R9M|A_R10M|A_R11M|A_R12M|A_R14M)
 
 #define emith_sh2_drc_exit() \
-       EOP_LDMFD_SP(A_R4M|A_R5M|A_R6M|A_R7M|A_R8M|A_R9M|A_R10M|A_R11M|A_R15M)
+       EOP_LDMFD_SP(A_R4M|A_R5M|A_R6M|A_R7M|A_R8M|A_R9M|A_R10M|A_R11M|A_R12M|A_R15M)
+
+#define emith_sh2_wcall(a, tab, ret_ptr) { \
+       int val_ = (char *)(ret_ptr) - (char *)tcache_ptr - 2*4; \
+       if (val_ >= 0) \
+               emith_add_r_r_imm(14, 15, val_); \
+       else if (val_ < 0) \
+               emith_sub_r_r_imm(14, 15, -val_); \
+       emith_lsr(12, a, SH2_WRITE_SHIFT); \
+       EOP_LDR_REG_LSL(A_COND_AL,12,tab,12,2); \
+       emith_ctx_read(2, offsetof(SH2, is_slave)); \
+       emith_jump_reg(12); \
+}
 
 #define emith_sh2_dtbf_loop() { \
        int cr, rn;                                                          \