faster effective_to_physical_address

Author: seppel

ChangeLog

@@ -1,13 +1,15 @@
  PearPC ChangeLog
 ==================
 
-version 0.5pre (not yet released):
+version 0.6pre (not yet released):
+
+version 0.5 (released 13. July 2011):
 	- slightly better code for lmw/stmw opcodes
 	- use "GNU as" instead of nasm
 	- use -mdynamic-no-pic on *darwin
-	- fixed strict-aliasing problems
+	- fixed some strict-aliasing problems
 	- JITC: AMD64 support
-	- compiles with gcc 4.3
+	- compiles with gcc 4.3, 4.4, 4.5
 
 version 0.4 (released 20. December 2005):
 	- CPU: stfiwx implemented

Makefile.am

@@ -5,7 +5,7 @@ SUBDIRS		= doc src scripts
 
 dist2: distdir
 	-chmod -R a+r $(distdir)
-	BZIP2=$(BZIP2_ENV) $(TAR) chojf $(distdir).tar.bz2 $(distdir)
+	BZIP2=$(BZIP2_ENV) $(AMTAR) chojf $(distdir).tar.bz2 $(distdir)
 	-rm -rf $(distdir)
 
 dist-all2: dist2

src/cpu/cpu_jitc_x86_64/jitc_mmu.S

@@ -154,17 +154,24 @@ ppc_pte_protection:
 
 damageframebuffer:
 	push	%rax
+	push	%rbx
 	push	%rcx
+	push	%rdx
 	push	%rsi
 	push	%rdi
 	push	%r8
 	mov	%edi, %ebx
 	sub	%edi, IO_GCARD_FRAMEBUFFER_PA_START
 	call	EXTERN(damageFrameBufferExt)
+	mov	%edi, %ebx
+	sub	%edi, IO_GCARD_FRAMEBUFFER_PA_START-4095
+	call	EXTERN(damageFrameBufferExt)
 	pop	%r8
 	pop	%rdi
 	pop	%rsi
+	pop	%rdx
 	pop	%rcx
+	pop	%rbx
 	pop	%rax
 	ret
 
@@ -214,10 +221,11 @@ damageframebuffer:
 	ret	8;                                                             \
 4:                                                                             \
 	cmp	%edx, IO_GCARD_FRAMEBUFFER_PA_START;                           \
-	jmp	5f;                                                            \
-/*	jb	5f;*/                                                          \
+	jb	5f;                                                            \
 	cmp	%edx, IO_GCARD_FRAMEBUFFER_PA_END;                             \
 	ja	5f;                                                            \
+	mov	%ebx, %edx;                                                    \
+	call	damageframebuffer;                                             \
 	sub	%edx, IO_GCARD_FRAMEBUFFER_PA_START;                           \
 	sub	%eax, IO_GCARD_FRAMEBUFFER_PA_START;                           \
 	add	%rdx, [EXTERN(gFrameBuffer)];                                  \
@@ -236,6 +244,7 @@ damageframebuffer:
 ##	param2: 0 for read, 8 for write
 ##	param3: data / code
 ##	param4: pt offset
+##	r9d     effective address
 #define pg_table_lookup(n, rw, datacode, offset)                               \
 	mov	%rax, [EXTERN(gMemory)];                                       \
 	mov	%eax, [%rax + %rbx + offset];                                  \
@@ -282,8 +291,9 @@ damageframebuffer:
 .if n==0;                                                                      \
 /*	add	%rsp, 8;	/* hash1, no longer needed */                  \
 	pop	%rdx;                                                          \
+.else;                                                                        \
 .endif;                                                                        \
-	pop	%rax;		/* the effective address */                    \
+	mov	%eax, %r9d; /* the effective address */                    \
 	jne	protection_fault_##rw##_##datacode;                            \
 	                                                                       \
 	/* 	update R and C bits */                                         \
@@ -322,8 +332,7 @@ damageframebuffer:
 .if datacode==data;                                                            \
 4:                                                                             \
 	cmp	%ebx, IO_GCARD_FRAMEBUFFER_PA_START;                           \
-/*	jb	5f;*/                                                          \
-	jmp	5f;                                                            \
+	jb	5f;                                                            \
 	cmp	%ebx, IO_GCARD_FRAMEBUFFER_PA_END;                             \
 	ja	5f;                                                            \
 	call	damageframebuffer;                                             \
@@ -441,7 +450,7 @@ EXPORT(ppc_effective_to_physical_code):
 
 	# %ebx = hash1
 
-	push	%rax
+	mov	%r9d, %eax
 	push	%rbx			# das brauch ich
 
 	and	%ebx, [curCPU(pagetable_hashmask)]
@@ -477,7 +486,7 @@ EXPORT(ppc_effective_to_physical_code):
 
 	# page lookup failed --> throw exception
 
-	pop	%rax
+	mov	%eax, %r9d
 	pop	%rdx			# return address is no longer needed
 	pop	%rcx			# bytes to roll back
 	add	%rsp, %rcx
@@ -555,7 +564,7 @@ ppc_effective_to_physical_data_read:
 
 	# %ebx = hash1
 
-	push	%rax
+	mov	%r9d, %eax
 	push	%rbx			# das brauch ich
 
 	and	%ebx, [curCPU(pagetable_hashmask)]
@@ -591,7 +600,7 @@ ppc_effective_to_physical_data_read:
 
 	# page lookup failed --> throw exception
 
-	pop	%rax
+	mov	%eax, %r9d
 	pop	%rdx			# return address is no longer needed
 	pop	%rcx			# bytes to roll back
 	add	%rsp, %rcx
@@ -662,7 +671,7 @@ ppc_effective_to_physical_data_write:
 
 	# %ebx = hash1
 
-	push	%rax
+	mov	%r9d, %eax
 	push	%rbx			# das brauch ich
 
 	and	%ebx, [curCPU(pagetable_hashmask)]
@@ -699,7 +708,7 @@ ppc_effective_to_physical_data_write:
 
 	# page lookup failed --> throw exception
 
-	pop	%rax
+	mov	%eax, %r9d
 	pop	%rdx			# return address is no longer needed
 	pop	%rbx			# bytes to roll back
 	add	%rsp, %rbx
@@ -723,7 +732,8 @@ EXPORT(ppc_write_effective_byte_asm):
 	push	%rdx
 	push	16			# roll back 16 bytes in case of exception
 	call	ppc_effective_to_physical_data_write
-	pop	%rdx
+	mov	%edx, [%rsp]
+	lea	%rsp, [%rsp+8]
 	jc	1f
 	mov	[%rax], %dl
 	ret
@@ -753,7 +763,8 @@ EXPORT(ppc_write_effective_half_asm):
 	push	%rdx
 	push	16			# roll back 16 bytes in case of exception
 	call	ppc_effective_to_physical_data_write
-	pop	%rdx
+	mov	%edx, [%rsp]
+	lea	%rsp, [%rsp+8]
 	jc	2f
 	mov	[%rax], %dh
 	mov	[%rax+1], %dl
@@ -818,7 +829,8 @@ EXPORT(ppc_write_effective_word_asm):
 	push	%rdx
 	push	16			# roll back 16 bytes in case of exception
 	call	ppc_effective_to_physical_data_write
-	pop	%rdx
+	mov	%edx, [%rsp]
+	lea	%rsp, [%rsp+8]
 	jc	2f
 	mov	[%rax], %edx
 	ret
@@ -925,7 +937,8 @@ EXPORT(ppc_write_effective_dword_asm):
 	push	%rdx
 	push	16			# roll back 16 bytes in case of exception
 	call	ppc_effective_to_physical_data_write
-	pop	%rdx
+	mov	%rdx, [%rsp]
+	lea	%rsp, [%rsp+8]
 	jc	2f
 	mov	[%rax], %rdx
 	ret

src/cpu/cpu_jitc_x86_64/ppc_mmu.cc

@@ -3066,6 +3066,7 @@ JITCFlow ppc_opc_gen_stfd(JITC &jitc)
 
 	jitc.asmResolveFixup(fixup2, jitc.asmHERE());
 #else
+	jitc.asmALU32(X86_MOV, RSI, jitc.pc);
 	jitc.asmCALL((NativeAddress)ppc_write_effective_dword_asm);
 #endif
 	return flowEndBlock;