drc: arm: use movw/movt
[picodrive.git] / cpu / drc / emit_arm.c
index d12db86..eb5f332 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_MSR_IMM(ror2,imm) EOP_C_MSR_IMM(A_COND_AL,ror2,imm)
 #define EOP_MSR_REG(rm)       EOP_C_MSR_REG(A_COND_AL,rm)
 
+#define EOP_MOVW(rd,imm) \
+       EMIT(0xe3000000 | ((rd)<<12) | ((imm)&0xfff) | (((imm)<<4)&0xf0000))
+
+#define EOP_MOVT(rd,imm) \
+       EMIT(0xe3400000 | ((rd)<<12) | (((imm)>>16)&0xfff) | (((imm)>>12)&0xf0000))
 
 // 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)
@@ -254,6 +262,19 @@ static void emith_op_imm2(int cond, int s, int op, int rd, int rn, unsigned int
                        imm = ~imm;
                        op = A_OP_MVN;
                }
+#ifdef HAVE_ARMV7
+               for (v = imm, ror2 = 0; v && !(v & 3); v >>= 2)
+                       ror2--;
+               if (v >> 8) {
+                       /* 2+ insns needed - prefer movw/movt */
+                       if (op == A_OP_MVN)
+                               imm = ~imm;
+                       EOP_MOVW(rd, imm);
+                       if (imm & 0xffff0000)
+                               EOP_MOVT(rd, imm);
+                       return;
+               }
+#endif
                break;
 
        case A_OP_EOR:
@@ -332,8 +353,8 @@ 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) { \
@@ -360,6 +381,9 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_mvn_r_r(d, s) \
        EOP_MVN_REG(A_COND_AL,0,d,s,A_AM1_LSL,0)
 
+#define emith_add_r_r_r_lsl(d, s1, s2, lslimm) \
+       EOP_ADD_REG(A_COND_AL,0,d,s1,s2,A_AM1_LSL,lslimm)
+
 #define emith_or_r_r_r_lsl(d, s1, s2, lslimm) \
        EOP_ORR_REG(A_COND_AL,0,d,s1,s2,A_AM1_LSL,lslimm)
 
@@ -375,6 +399,9 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_eor_r_r_lsr(d, s, lsrimm) \
        emith_eor_r_r_r_lsr(d, d, s, lsrimm)
 
+#define emith_add_r_r_r(d, s1, s2) \
+       emith_add_r_r_r_lsl(d, s1, s2, 0)
+
 #define emith_or_r_r_r(d, s1, s2) \
        emith_or_r_r_r_lsl(d, s1, s2, 0)
 
@@ -382,7 +409,7 @@ static int emith_xbranch(int cond, void *target, int is_call)
        emith_eor_r_r_r_lsl(d, s1, s2, 0)
 
 #define emith_add_r_r(d, s) \
-       EOP_ADD_REG(A_COND_AL,0,d,d,s,A_AM1_LSL,0)
+       emith_add_r_r_r(d, d, s)
 
 #define emith_sub_r_r(d, s) \
        EOP_SUB_REG(A_COND_AL,0,d,d,s,A_AM1_LSL,0)
@@ -630,8 +657,20 @@ 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
+#define emith_do_caller_regs(mask, func) { \
+       u32 _reg_mask = (mask) & 0x500f; \
+       if (_reg_mask) { \
+               if (__builtin_parity(_reg_mask) == 1) \
+                       _reg_mask |= 0x10; /* eabi align */ \
+               func(_reg_mask); \
+       } \
+}
+
+#define emith_save_caller_regs(mask) \
+       emith_do_caller_regs(mask, EOP_STMFD_SP)
+
+#define emith_restore_caller_regs(mask) \
+       emith_do_caller_regs(mask, EOP_LDMFD_SP)
 
 // upto 4 args
 #define emith_pass_arg_r(arg, reg) \
@@ -658,6 +697,11 @@ static int emith_xbranch(int cond, void *target, int is_call)
        *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)
 
@@ -690,22 +734,30 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #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)
-
-#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_); \
+       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) { \
        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_move_r_r(2, CONTEXT_REG); \
        emith_jump_reg(12); \
 }