Make ewram memory lineal

This saves a few cycles in MIPS and simplifies a bit the core.
Removed the write map, only affects interpreter performance very
minimally. Rewired ARM and x86 handlers to support direct access to
I/EWRAM (and VRAM on ARM) to compensate. Overall performance is slightly
better but code is cleaner and allows for further improvements in the
dynarecs.

2 files changed, 61 insertions, 81 deletions

diff --git a/x86/x86_emit.h b/x86/x86_emit.h
index 67a3dc2..68930e1 100644
--- a/x86/x86_emit.h
+++ b/x86/x86_emit.h
@@ -1485,23 +1485,13 @@ u32 function_cc execute_aligned_load32(u32 address)
     return read_memory32(address);
 }
 
-void function_cc execute_aligned_store32(u32 address, u32 source)
-{
-  u8 *map;
-
-  if(!(address & 0xF0000000) && (map = memory_map_write[address >> 15]))
-    address32(map, address & 0x7FFF) = source;
-  else
-    write_memory32(address, source);
-}
-
 #define arm_block_memory_load()                                               \
   generate_function_call(execute_aligned_load32);                             \
   generate_store_reg(rv, i)                                                   \
 
 #define arm_block_memory_store()                                              \
   generate_load_reg_pc(a1, i, 8);                                             \
-  generate_function_call(execute_aligned_store32)                             \
+  generate_function_call(write_memory32)                                      \
 
 #define arm_block_memory_final_load()                                         \
   arm_block_memory_load()                                                     \
@@ -1956,7 +1946,7 @@ u32 function_cc execute_ror_imm_op(u32 value, u32 shift)
 #define thumb_block_memory_extra_push_lr(base_reg)                            \
   generate_add_reg_reg_imm(a0, s0, (bit_count[reg_list] * 4));                \
   generate_load_reg(a1, REG_LR);                                              \
-  generate_function_call(execute_aligned_store32)                             \
+  generate_function_call(write_memory32)                                      \
 
 #define thumb_block_memory_load()                                             \
   generate_function_call(execute_aligned_load32);                             \
@@ -1964,7 +1954,7 @@ u32 function_cc execute_ror_imm_op(u32 value, u32 shift)
 
 #define thumb_block_memory_store()                                            \
   generate_load_reg(a1, i);                                                   \
-  generate_function_call(execute_aligned_store32)                             \
+  generate_function_call(write_memory32)                                      \
 
 #define thumb_block_memory_final_load()                                       \
   thumb_block_memory_load()                                                   \
diff --git a/x86/x86_stub.S b/x86/x86_stub.S
index 1e338a4..595a789 100644
--- a/x86/x86_stub.S
+++ b/x86/x86_stub.S
@@ -29,7 +29,6 @@
 #define _execute_store_cpsr execute_store_cpsr
 #define _execute_arm_translate execute_arm_translate
 #define _memory_map_read memory_map_read
-#define _memory_map_write memory_map_write
 #define _reg reg
 #define _reg_mode reg_mode
 #define _oam_update oam_update
@@ -68,7 +67,6 @@
 .global _execute_arm_translate
 
 .global _memory_map_read
-.global _memory_map_write
 .global _reg
 .global _reg_mode
 .global _spsr
@@ -231,6 +229,20 @@ ext_store_eeprom:
 
 # 8bit ext memory routines
 
+ext_store_iwram8:
+  and $0x7FFF, %eax                 # wrap around address
+  mov %dl, (_iwram+0x8000)(%eax)    # perform store
+  cmpb $0, _iwram(%eax)             # Check SMC mirror
+  jne smc_write
+  ret
+
+ext_store_ewram8:
+  and $0x3FFFF, %eax                # wrap around address
+  mov %dl, _ewram(%eax)             # perform store
+  cmpb $0, (_ewram+0x40000)(%eax)   # Check SMC mirror
+  jne smc_write
+  ret
+
 ext_store_io8:
   and $0x3FF, %eax            # wrap around address
   and $0xFF, %edx
@@ -267,8 +279,8 @@ ext_store_backup:
 ext_store_u8_jtable:
   .long ext_store_ignore      # 0x00 BIOS, ignore
   .long ext_store_ignore      # 0x01 invalid, ignore
-  .long ext_store_ignore      # 0x02 EWRAM, should have been hit already
-  .long ext_store_ignore      # 0x03 IWRAM, should have been hit already
+  .long ext_store_ewram8      # 0x02 EWRAM
+  .long ext_store_iwram8      # 0x03 IWRAM
   .long ext_store_io8         # 0x04 I/O registers
   .long ext_store_palette8    # 0x05 Palette RAM
   .long ext_store_vram8       # 0x06 VRAM
@@ -281,7 +293,12 @@ ext_store_u8_jtable:
   .long ext_store_eeprom      # 0x0D EEPROM (possibly)
   .long ext_store_backup      # 0x0E Flash ROM/SRAM
 
-ext_store_u8:
+# eax: address to write to
+# edx: value to write
+# ecx: current pc
+
+_execute_store_u8:
+  mov %ecx, REG_PC(%ebx)      # write out the PC
   mov %eax, %ecx              # ecx = address
   shr $24, %ecx               # ecx = address >> 24
   cmp $15, %ecx
@@ -290,46 +307,21 @@ ext_store_u8:
   mov ext_store_u8_jtable(, %ecx, 4), %ecx
   jmp *%ecx                   # jump to table index
 
-# eax: address to write to
-# edx: value to write
-# ecx: current pc
+# 16bit ext memory routines
 
-_execute_store_u8:
-  mov %ecx, REG_PC(%ebx)      # write out the PC
-  mov %eax, %ecx              # ecx = address
-  test $0xF0000000, %ecx      # check address range
-  jnz ext_store_u8            # if above perform an extended write
-  shr $15, %ecx               # ecx = page number of address
-  # load the corresponding memory map offset
-  mov _memory_map_write(, %ecx, 4), %ecx
-  test %ecx, %ecx             # see if it's NULL
-  jz ext_store_u8             # if so perform an extended write
-  and $0x7FFF, %eax           # isolate the lower 15bits of the address
-  mov %dl, (%eax, %ecx)       # store the value
-  # check for self-modifying code
-  testb $0xFF, -32768(%eax, %ecx)
+ext_store_iwram16:
+  and $0x7FFF, %eax                 # wrap around address
+  mov %dx, (_iwram+0x8000)(%eax)    # perform store
+  cmpw $0, _iwram(%eax)             # Check SMC mirror
   jne smc_write
-  ret                         # return
+  ret
 
-_execute_store_u16:
-  mov %ecx, REG_PC(%ebx)      # write out the PC
-  and $~0x01, %eax            # fix alignment
-  mov %eax, %ecx              # ecx = address
-  test $0xF0000000, %ecx      # check address range
-  jnz ext_store_u16           # if above perform an extended write
-  shr $15, %ecx               # ecx = page number of address
-  # load the corresponding memory map offset
-  mov _memory_map_write(, %ecx, 4), %ecx
-  test %ecx, %ecx             # see if it's NULL
-  jz ext_store_u16            # if so perform an extended write
-  and $0x7FFF, %eax           # isolate the lower 15bits of the address
-  mov %dx, (%eax, %ecx)       # store the value
-  # check for self-modifying code
-  testw $0xFFFF, -32768(%eax, %ecx)
+ext_store_ewram16:
+  and $0x3FFFF, %eax                # wrap around address
+  mov %dx, _ewram(%eax)             # perform store
+  cmpw $0, (_ewram+0x40000)(%eax)   # Check SMC mirror
   jne smc_write
-  ret                         # return
-
-# 16bit ext memory routines
+  ret
 
 ext_store_io16:
   and $0x3FF, %eax            # wrap around address
@@ -377,8 +369,8 @@ ext_store_rtc:
 ext_store_u16_jtable:
   .long ext_store_ignore      # 0x00 BIOS, ignore
   .long ext_store_ignore      # 0x01 invalid, ignore
-  .long ext_store_ignore      # 0x02 EWRAM, should have been hit already
-  .long ext_store_ignore      # 0x03 IWRAM, should have been hit already
+  .long ext_store_ewram16     # 0x02 EWRAM
+  .long ext_store_iwram16     # 0x03 IWRAM
   .long ext_store_io16        # 0x04 I/O registers
   .long ext_store_palette16   # 0x05 Palette RAM
   .long ext_store_vram16      # 0x06 VRAM
@@ -391,7 +383,9 @@ ext_store_u16_jtable:
   .long ext_store_eeprom      # 0x0D EEPROM (possibly)
   .long ext_store_ignore      # 0x0E Flash ROM/SRAM must be 8bit
 
-ext_store_u16:
+_execute_store_u16:
+  mov %ecx, REG_PC(%ebx)      # write out the PC
+  and $~0x01, %eax            # fix alignment
   mov %eax, %ecx              # ecx = address
   shr $24, %ecx               # ecx = address >> 24
   cmp $15, %ecx
@@ -400,25 +394,21 @@ ext_store_u16:
   mov ext_store_u16_jtable(, %ecx, 4), %ecx
   jmp *%ecx                   # jump to table index
 
-_execute_store_u32:
-  mov %ecx, REG_PC(%ebx)      # write out the PC
-  and $~0x03, %eax            # fix alignment
-  mov %eax, %ecx              # ecx = address
-  test $0xF0000000, %ecx      # check address range
-  jnz ext_store_u32           # if above perform an extended write
-  shr $15, %ecx               # ecx = page number of address
-  # load the corresponding memory map offset
-  mov _memory_map_write(, %ecx, 4), %ecx
-  test %ecx, %ecx             # see if it's NULL
-  jz ext_store_u32            # if so perform an extended write
-  and $0x7FFF, %eax           # isolate the lower 15bits of the address
-  mov %edx, (%eax, %ecx)      # store the value
-  # check for self-modifying code
-  testl $0xFFFFFFFF, -32768(%eax, %ecx)
+# 32bit ext memory routines
+
+ext_store_iwram32:
+  and $0x7FFF, %eax                 # wrap around address
+  mov %edx, (_iwram+0x8000)(%eax)   # perform store
+  cmpl $0, _iwram(%eax)             # Check SMC mirror
   jne smc_write
-  ret                         # return it
+  ret
 
-# 32bit ext memory routines
+ext_store_ewram32:
+  and $0x3FFFF, %eax                # wrap around address
+  mov %edx, _ewram(%eax)            # perform store
+  cmpl $0, (_ewram+0x40000)(%eax)   # Check SMC mirror
+  jne smc_write
+  ret
 
 ext_store_io32:
   and $0x3FF, %eax            # wrap around address
@@ -451,8 +441,8 @@ ext_store_oam32:
 ext_store_u32_jtable:
   .long ext_store_ignore      # 0x00 BIOS, ignore
   .long ext_store_ignore      # 0x01 invalid, ignore
-  .long ext_store_ignore      # 0x02 EWRAM, should have been hit already
-  .long ext_store_ignore      # 0x03 IWRAM, should have been hit already
+  .long ext_store_ewram32     # 0x02 EWRAM
+  .long ext_store_iwram32     # 0x03 IWRAM
   .long ext_store_io32        # 0x04 I/O registers
   .long ext_store_palette32   # 0x05 Palette RAM
   .long ext_store_vram32      # 0x06 VRAM
@@ -466,7 +456,9 @@ ext_store_u32_jtable:
   .long ext_store_ignore      # 0x0E Flash ROM/SRAM must be 8bit
 
 
-ext_store_u32:
+_execute_store_u32:
+  mov %ecx, REG_PC(%ebx)      # write out the PC
+  and $~0x03, %eax            # fix alignment
   mov %eax, %ecx              # ecx = address
   shr $24, %ecx               # ecx = address >> 24
   cmp $15, %ecx
@@ -507,8 +499,8 @@ smc_write:
   call _flush_translation_cache_ram
 
 lookup_pc:
-  add $4, %esp
-  movl $0, CHANGED_PC_STATUS(%ebx)
+  add $4, %esp                       # Can't return, discard addr
+  movl $0, CHANGED_PC_STATUS(%ebx)   # Lookup new block and jump to it
   mov REG_PC(%ebx), %eax
   testl $0x20, REG_CPSR(%ebx)
   jz lookup_pc_arm
@@ -577,7 +569,5 @@ _reg_mode:
 
 _memory_map_read:
   .space 0x8000
-_memory_map_write:
-  .space 0x8000