summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--configs/default2
-rw-r--r--src/gallium/SConscript1
-rw-r--r--src/gallium/auxiliary/rtasm/Makefile20
-rw-r--r--src/gallium/auxiliary/rtasm/SConscript11
-rw-r--r--src/gallium/auxiliary/rtasm/execmem.c133
-rw-r--r--src/gallium/auxiliary/rtasm/execmem.h45
-rw-r--r--src/gallium/auxiliary/rtasm/mm.c283
-rw-r--r--src/gallium/auxiliary/rtasm/mm.h89
-rw-r--r--src/gallium/auxiliary/rtasm/x86sse.c1195
-rw-r--r--src/gallium/auxiliary/rtasm/x86sse.h256
10 files changed, 2034 insertions, 1 deletions
diff --git a/configs/default b/configs/default
index 7f659725cb..c9be5ec3e3 100644
--- a/configs/default
+++ b/configs/default
@@ -68,7 +68,7 @@ PROGRAM_DIRS = demos redbook samples glsl xdemos
# Gallium directories and
-GALLIUM_AUXILIARY_DIRS = draw cso_cache pipebuffer tgsi util
+GALLIUM_AUXILIARY_DIRS = draw cso_cache pipebuffer tgsi rtasm util
GALLIUM_AUXILIARIES = $(foreach DIR,$(GALLIUM_AUXILIARY_DIRS),$(TOP)/src/gallium/auxiliary/$(DIR)/lib$(DIR).a)
GALLIUM_DRIVER_DIRS = softpipe i915simple i965simple failover
GALLIUM_DRIVERS = $(foreach DIR,$(GALLIUM_DRIVER_DIRS),$(TOP)/src/gallium/drivers/$(DIR)/lib$(DIR).a)
diff --git a/src/gallium/SConscript b/src/gallium/SConscript
index c505eee792..a08b4b830e 100644
--- a/src/gallium/SConscript
+++ b/src/gallium/SConscript
@@ -10,6 +10,7 @@ Export('auxiliaries')
SConscript([
# NOTE: order matters!
'auxiliary/util/SConscript',
+ 'auxiliary/rtasm/SConscript',
'auxiliary/tgsi/SConscript',
'auxiliary/cso_cache/SConscript',
'auxiliary/draw/SConscript',
diff --git a/src/gallium/auxiliary/rtasm/Makefile b/src/gallium/auxiliary/rtasm/Makefile
new file mode 100644
index 0000000000..b3b9934e10
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/Makefile
@@ -0,0 +1,20 @@
+
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = rtasm
+
+DRIVER_SOURCES = \
+ x86sse.c \
+ mm.c \
+ execmem.c
+
+C_SOURCES = \
+ $(DRIVER_SOURCES)
+
+ASM_SOURCES =
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/rtasm/SConscript b/src/gallium/auxiliary/rtasm/SConscript
new file mode 100644
index 0000000000..c5b1551786
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/SConscript
@@ -0,0 +1,11 @@
+Import('*')
+
+rtasm = env.ConvenienceLibrary(
+ target = 'rtasm',
+ source = [
+ 'x86sse.c',
+ 'mm.c',
+ 'execmem.c',
+ ])
+
+auxiliaries.insert(0, rtasm)
diff --git a/src/gallium/auxiliary/rtasm/execmem.c b/src/gallium/auxiliary/rtasm/execmem.c
new file mode 100644
index 0000000000..c7c35f7ef2
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/execmem.c
@@ -0,0 +1,133 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005 Brian Paul All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+/**
+ * \file exemem.c
+ * Functions for allocating executable memory.
+ *
+ * \author Keith Whitwell
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_thread.h"
+
+#include "execmem.h"
+
+
+#if defined(__linux__)
+
+/*
+ * Allocate a large block of memory which can hold code then dole it out
+ * in pieces by means of the generic memory manager code.
+*/
+
+#include <unistd.h>
+#include <sys/mman.h>
+#include "mm.h"
+
+#define EXEC_HEAP_SIZE (10*1024*1024)
+
+_glthread_DECLARE_STATIC_MUTEX(exec_mutex);
+
+static struct mem_block *exec_heap = NULL;
+static unsigned char *exec_mem = NULL;
+
+
+static void
+init_heap(void)
+{
+ if (!exec_heap)
+ exec_heap = mmInit( 0, EXEC_HEAP_SIZE );
+
+ if (!exec_mem)
+ exec_mem = (unsigned char *) mmap(0, EXEC_HEAP_SIZE,
+ PROT_EXEC | PROT_READ | PROT_WRITE,
+ MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+}
+
+
+void *
+_mesa_exec_malloc(size_t size)
+{
+ struct mem_block *block = NULL;
+ void *addr = NULL;
+
+ _glthread_LOCK_MUTEX(exec_mutex);
+
+ init_heap();
+
+ if (exec_heap) {
+ size = (size + 31) & ~31;
+ block = mmAllocMem( exec_heap, size, 32, 0 );
+ }
+
+ if (block)
+ addr = exec_mem + block->ofs;
+ else
+ debug_printf("_mesa_exec_malloc failed\n");
+
+ _glthread_UNLOCK_MUTEX(exec_mutex);
+
+ return addr;
+}
+
+
+void
+_mesa_exec_free(void *addr)
+{
+ _glthread_LOCK_MUTEX(exec_mutex);
+
+ if (exec_heap) {
+ struct mem_block *block = mmFindBlock(exec_heap, (unsigned char *)addr - exec_mem);
+
+ if (block)
+ mmFreeMem(block);
+ }
+
+ _glthread_UNLOCK_MUTEX(exec_mutex);
+}
+
+
+#else
+
+/*
+ * Just use regular memory.
+ */
+
+void *
+_mesa_exec_malloc(GLuint size)
+{
+ return _mesa_malloc( size );
+}
+
+
+void
+_mesa_exec_free(void *addr)
+{
+ _mesa_free(addr);
+}
+
+
+#endif
diff --git a/src/gallium/auxiliary/rtasm/execmem.h b/src/gallium/auxiliary/rtasm/execmem.h
new file mode 100644
index 0000000000..9fd4569165
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/execmem.h
@@ -0,0 +1,45 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005 Brian Paul All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file exemem.c
+ * Functions for allocating executable memory.
+ *
+ * \author Keith Whitwell
+ */
+
+#ifndef _EXECMEM_H_
+#define _EXECMEM_H_
+
+#include "pipe/p_compiler.h"
+
+
+extern void *
+_mesa_exec_malloc( size_t size );
+
+
+extern void
+_mesa_exec_free( void *addr );
+
+
+#endif
diff --git a/src/gallium/auxiliary/rtasm/mm.c b/src/gallium/auxiliary/rtasm/mm.c
new file mode 100644
index 0000000000..15f50491da
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/mm.c
@@ -0,0 +1,283 @@
+/*
+ * GLX Hardware Device Driver common code
+ * Copyright (C) 1999 Wittawat Yamwong
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * WITTAWAT YAMWONG, OR ANY OTHER CONTRIBUTORS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE
+ * OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_util.h"
+#include "pipe/p_debug.h"
+
+#include "mm.h"
+
+
+void
+mmDumpMemInfo(const struct mem_block *heap)
+{
+ debug_printf("Memory heap %p:\n", (void *)heap);
+ if (heap == 0) {
+ debug_printf(" heap == 0\n");
+ } else {
+ const struct mem_block *p;
+
+ for(p = heap->next; p != heap; p = p->next) {
+ debug_printf(" Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
+ p->free ? 'F':'.',
+ p->reserved ? 'R':'.');
+ }
+
+ debug_printf("\nFree list:\n");
+
+ for(p = heap->next_free; p != heap; p = p->next_free) {
+ debug_printf(" FREE Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
+ p->free ? 'F':'.',
+ p->reserved ? 'R':'.');
+ }
+
+ }
+ debug_printf("End of memory blocks\n");
+}
+
+struct mem_block *
+mmInit(int ofs, int size)
+{
+ struct mem_block *heap, *block;
+
+ if (size <= 0)
+ return NULL;
+
+ heap = CALLOC_STRUCT(mem_block);
+ if (!heap)
+ return NULL;
+
+ block = CALLOC_STRUCT(mem_block);
+ if (!block) {
+ FREE(heap);
+ return NULL;
+ }
+
+ heap->next = block;
+ heap->prev = block;
+ heap->next_free = block;
+ heap->prev_free = block;
+
+ block->heap = heap;
+ block->next = heap;
+ block->prev = heap;
+ block->next_free = heap;
+ block->prev_free = heap;
+
+ block->ofs = ofs;
+ block->size = size;
+ block->free = 1;
+
+ return heap;
+}
+
+
+static struct mem_block *
+SliceBlock(struct mem_block *p,
+ int startofs, int size,
+ int reserved, int alignment)
+{
+ struct mem_block *newblock;
+
+ /* break left [p, newblock, p->next], then p = newblock */
+ if (startofs > p->ofs) {
+ newblock = CALLOC_STRUCT(mem_block);
+ if (!newblock)
+ return NULL;
+ newblock->ofs = startofs;
+ newblock->size = p->size - (startofs - p->ofs);
+ newblock->free = 1;
+ newblock->heap = p->heap;
+
+ newblock->next = p->next;
+ newblock->prev = p;
+ p->next->prev = newblock;
+ p->next = newblock;
+
+ newblock->next_free = p->next_free;
+ newblock->prev_free = p;
+ p->next_free->prev_free = newblock;
+ p->next_free = newblock;
+
+ p->size -= newblock->size;
+ p = newblock;
+ }
+
+ /* break right, also [p, newblock, p->next] */
+ if (size < p->size) {
+ newblock = CALLOC_STRUCT(mem_block);
+ if (!newblock)
+ return NULL;
+ newblock->ofs = startofs + size;
+ newblock->size = p->size - size;
+ newblock->free = 1;
+ newblock->heap = p->heap;
+
+ newblock->next = p->next;
+ newblock->prev = p;
+ p->next->prev = newblock;
+ p->next = newblock;
+
+ newblock->next_free = p->next_free;
+ newblock->prev_free = p;
+ p->next_free->prev_free = newblock;
+ p->next_free = newblock;
+
+ p->size = size;
+ }
+
+ /* p = middle block */
+ p->free = 0;
+
+ /* Remove p from the free list:
+ */
+ p->next_free->prev_free = p->prev_free;
+ p->prev_free->next_free = p->next_free;
+
+ p->next_free = 0;
+ p->prev_free = 0;
+
+ p->reserved = reserved;
+ return p;
+}
+
+
+struct mem_block *
+mmAllocMem(struct mem_block *heap, int size, int align2, int startSearch)
+{
+ struct mem_block *p;
+ const int mask = (1 << align2)-1;
+ int startofs = 0;
+ int endofs;
+
+ if (!heap || align2 < 0 || size <= 0)
+ return NULL;
+
+ for (p = heap->next_free; p != heap; p = p->next_free) {
+ assert(p->free);
+
+ startofs = (p->ofs + mask) & ~mask;
+ if ( startofs < startSearch ) {
+ startofs = startSearch;
+ }
+ endofs = startofs+size;
+ if (endofs <= (p->ofs+p->size))
+ break;
+ }
+
+ if (p == heap)
+ return NULL;
+
+ assert(p->free);
+ p = SliceBlock(p,startofs,size,0,mask+1);
+
+ return p;
+}
+
+
+struct mem_block *
+mmFindBlock(struct mem_block *heap, int start)
+{
+ struct mem_block *p;
+
+ for (p = heap->next; p != heap; p = p->next) {
+ if (p->ofs == start)
+ return p;
+ }
+
+ return NULL;
+}
+
+
+static INLINE int
+Join2Blocks(struct mem_block *p)
+{
+ /* XXX there should be some assertions here */
+
+ /* NOTE: heap->free == 0 */
+
+ if (p->free && p->next->free) {
+ struct mem_block *q = p->next;
+
+ assert(p->ofs + p->size == q->ofs);
+ p->size += q->size;
+
+ p->next = q->next;
+ q->next->prev = p;
+
+ q->next_free->prev_free = q->prev_free;
+ q->prev_free->next_free = q->next_free;
+
+ FREE(q);
+ return 1;
+ }
+ return 0;
+}
+
+int
+mmFreeMem(struct mem_block *b)
+{
+ if (!b)
+ return 0;
+
+ if (b->free) {
+ debug_printf("block already free\n");
+ return -1;
+ }
+ if (b->reserved) {
+ debug_printf("block is reserved\n");
+ return -1;
+ }
+
+ b->free = 1;
+ b->next_free = b->heap->next_free;
+ b->prev_free = b->heap;
+ b->next_free->prev_free = b;
+ b->prev_free->next_free = b;
+
+ Join2Blocks(b);
+ if (b->prev != b->heap)
+ Join2Blocks(b->prev);
+
+ return 0;
+}
+
+
+void
+mmDestroy(struct mem_block *heap)
+{
+ struct mem_block *p;
+
+ if (!heap)
+ return;
+
+ for (p = heap->next; p != heap; ) {
+ struct mem_block *next = p->next;
+ FREE(p);
+ p = next;
+ }
+
+ FREE(heap);
+}
diff --git a/src/gallium/auxiliary/rtasm/mm.h b/src/gallium/auxiliary/rtasm/mm.h
new file mode 100644
index 0000000000..f469b18d3e
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/mm.h
@@ -0,0 +1,89 @@
+/*
+ * GLX Hardware Device Driver common code
+ * Copyright (C) 1999 Wittawat Yamwong
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * KEITH WHITWELL, OR ANY OTHER CONTRIBUTORS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE
+ * OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/**
+ * Memory manager code. Primarily used by device drivers to manage texture
+ * heaps, etc.
+ */
+
+
+#ifndef MM_H
+#define MM_H
+
+
+struct mem_block {
+ struct mem_block *next, *prev;
+ struct mem_block *next_free, *prev_free;
+ struct mem_block *heap;
+ int ofs,size;
+ unsigned int free:1;
+ unsigned int reserved:1;
+};
+
+
+
+/**
+ * input: total size in bytes
+ * return: a heap pointer if OK, NULL if error
+ */
+extern struct mem_block *mmInit(int ofs, int size);
+
+/**
+ * Allocate 'size' bytes with 2^align2 bytes alignment,
+ * restrict the search to free memory after 'startSearch'
+ * depth and back buffers should be in different 4mb banks
+ * to get better page hits if possible
+ * input: size = size of block
+ * align2 = 2^align2 bytes alignment
+ * startSearch = linear offset from start of heap to begin search
+ * return: pointer to the allocated block, 0 if error
+ */
+extern struct mem_block *mmAllocMem(struct mem_block *heap, int size, int align2,
+ int startSearch);
+
+/**
+ * Free block starts at offset
+ * input: pointer to a block
+ * return: 0 if OK, -1 if error
+ */
+extern int mmFreeMem(struct mem_block *b);
+
+/**
+ * Free block starts at offset
+ * input: pointer to a heap, start offset
+ * return: pointer to a block
+ */
+extern struct mem_block *mmFindBlock(struct mem_block *heap, int start);
+
+/**
+ * destroy MM
+ */
+extern void mmDestroy(struct mem_block *mmInit);
+
+/**
+ * For debuging purpose.
+ */
+extern void mmDumpMemInfo(const struct mem_block *mmInit);
+
+#endif
diff --git a/src/gallium/auxiliary/rtasm/x86sse.c b/src/gallium/auxiliary/rtasm/x86sse.c
new file mode 100644
index 0000000000..fff6f77a6b
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/x86sse.c
@@ -0,0 +1,1195 @@
+#if defined(__i386__) || defined(__386__)
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+
+#include "x86sse.h"
+
+#define DISASSEM 0
+#define X86_TWOB 0x0f
+
+static unsigned char *cptr( void (*label)() )
+{
+ return (unsigned char *)(unsigned long)label;
+}
+
+
+static void do_realloc( struct x86_function *p )
+{
+ if (p->size == 0) {
+ p->size = 1024;
+ p->store = _mesa_exec_malloc(p->size);
+ p->csr = p->store;
+ }
+ else {
+ unsigned used = p->csr - p->store;
+ unsigned char *tmp = p->store;
+ p->size *= 2;
+ p->store = _mesa_exec_malloc(p->size);
+ memcpy(p->store, tmp, used);
+ p->csr = p->store + used;
+ _mesa_exec_free(tmp);
+ }
+}
+
+/* Emit bytes to the instruction stream:
+ */
+static unsigned char *reserve( struct x86_function *p, int bytes )
+{
+ if (p->csr + bytes - p->store > p->size)
+ do_realloc(p);
+
+ {
+ unsigned char *csr = p->csr;
+ p->csr += bytes;
+ return csr;
+ }
+}
+
+
+
+static void emit_1b( struct x86_function *p, char b0 )
+{
+ char *csr = (char *)reserve(p, 1);
+ *csr = b0;
+}
+
+static void emit_1i( struct x86_function *p, int i0 )
+{
+ int *icsr = (int *)reserve(p, sizeof(i0));
+ *icsr = i0;
+}
+
+static void emit_1ub( struct x86_function *p, unsigned char b0 )
+{
+ unsigned char *csr = reserve(p, 1);
+ *csr++ = b0;
+}
+
+static void emit_2ub( struct x86_function *p, unsigned char b0, unsigned char b1 )
+{
+ unsigned char *csr = reserve(p, 2);
+ *csr++ = b0;
+ *csr++ = b1;
+}
+
+static void emit_3ub( struct x86_function *p, unsigned char b0, unsigned char b1, unsigned char b2 )
+{
+ unsigned char *csr = reserve(p, 3);
+ *csr++ = b0;
+ *csr++ = b1;
+ *csr++ = b2;
+}
+
+
+/* Build a modRM byte + possible displacement. No treatment of SIB
+ * indexing. BZZT - no way to encode an absolute address.
+ */
+static void emit_modrm( struct x86_function *p,
+ struct x86_reg reg,
+ struct x86_reg regmem )
+{
+ unsigned char val = 0;
+
+ assert(reg.mod == mod_REG);
+
+ val |= regmem.mod << 6; /* mod field */
+ val |= reg.idx << 3; /* reg field */
+ val |= regmem.idx; /* r/m field */
+
+ emit_1ub(p, val);
+
+ /* Oh-oh we've stumbled into the SIB thing.
+ */
+ if (regmem.file == file_REG32 &&
+ regmem.idx == reg_SP) {
+ emit_1ub(p, 0x24); /* simplistic! */
+ }
+
+ switch (regmem.mod) {
+ case mod_REG:
+ case mod_INDIRECT:
+ break;
+ case mod_DISP8:
+ emit_1b(p, regmem.disp);
+ break;
+ case mod_DISP32:
+ emit_1i(p, regmem.disp);
+ break;
+ default:
+ assert(0);
+ break;
+ }
+}
+
+
+static void emit_modrm_noreg( struct x86_function *p,
+ unsigned op,
+ struct x86_reg regmem )
+{
+ struct x86_reg dummy = x86_make_reg(file_REG32, op);
+ emit_modrm(p, dummy, regmem);
+}
+
+/* Many x86 instructions have two opcodes to cope with the situations
+ * where the destination is a register or memory reference
+ * respectively. This function selects the correct opcode based on
+ * the arguments presented.
+ */
+static void emit_op_modrm( struct x86_function *p,
+ unsigned char op_dst_is_reg,
+ unsigned char op_dst_is_mem,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ switch (dst.mod) {
+ case mod_REG:
+ emit_1ub(p, op_dst_is_reg);
+ emit_modrm(p, dst, src);
+ break;
+ case mod_INDIRECT:
+ case mod_DISP32:
+ case mod_DISP8:
+ assert(src.mod == mod_REG);
+ emit_1ub(p, op_dst_is_mem);
+ emit_modrm(p, src, dst);
+ break;
+ default:
+ assert(0);
+ break;
+ }
+}
+
+
+
+
+
+
+
+/* Create and manipulate registers and regmem values:
+ */
+struct x86_reg x86_make_reg( enum x86_reg_file file,
+ enum x86_reg_name idx )
+{
+ struct x86_reg reg;
+
+ reg.file = file;
+ reg.idx = idx;
+ reg.mod = mod_REG;
+ reg.disp = 0;
+
+ return reg;
+}
+
+struct x86_reg x86_make_disp( struct x86_reg reg,
+ int disp )
+{
+ assert(reg.file == file_REG32);
+
+ if (reg.mod == mod_REG)
+ reg.disp = disp;
+ else
+ reg.disp += disp;
+
+ if (reg.disp == 0)
+ reg.mod = mod_INDIRECT;
+ else if (reg.disp <= 127 && reg.disp >= -128)
+ reg.mod = mod_DISP8;
+ else
+ reg.mod = mod_DISP32;
+
+ return reg;
+}
+
+struct x86_reg x86_deref( struct x86_reg reg )
+{
+ return x86_make_disp(reg, 0);
+}
+
+struct x86_reg x86_get_base_reg( struct x86_reg reg )
+{
+ return x86_make_reg( reg.file, reg.idx );
+}
+
+unsigned char *x86_get_label( struct x86_function *p )
+{
+ return p->csr;
+}
+
+
+
+/***********************************************************************
+ * x86 instructions
+ */
+
+
+void x86_jcc( struct x86_function *p,
+ enum x86_cc cc,
+ unsigned char *label )
+{
+ int offset = label - (x86_get_label(p) + 2);
+
+ if (offset <= 127 && offset >= -128) {
+ emit_1ub(p, 0x70 + cc);
+ emit_1b(p, (char) offset);
+ }
+ else {
+ offset = label - (x86_get_label(p) + 6);
+ emit_2ub(p, 0x0f, 0x80 + cc);
+ emit_1i(p, offset);
+ }
+}
+
+/* Always use a 32bit offset for forward jumps:
+ */
+unsigned char *x86_jcc_forward( struct x86_function *p,
+ enum x86_cc cc )
+{
+ emit_2ub(p, 0x0f, 0x80 + cc);
+ emit_1i(p, 0);
+ return x86_get_label(p);
+}
+
+unsigned char *x86_jmp_forward( struct x86_function *p)
+{
+ emit_1ub(p, 0xe9);
+ emit_1i(p, 0);
+ return x86_get_label(p);
+}
+
+unsigned char *x86_call_forward( struct x86_function *p)
+{
+ emit_1ub(p, 0xe8);
+ emit_1i(p, 0);
+ return x86_get_label(p);
+}
+
+/* Fixup offset from forward jump:
+ */
+void x86_fixup_fwd_jump( struct x86_function *p,
+ unsigned char *fixup )
+{
+ *(int *)(fixup - 4) = x86_get_label(p) - fixup;
+}
+
+void x86_jmp( struct x86_function *p, unsigned char *label)
+{
+ emit_1ub(p, 0xe9);
+ emit_1i(p, label - x86_get_label(p) - 4);
+}
+
+#if 0
+/* This doesn't work once we start reallocating & copying the
+ * generated code on buffer fills, because the call is relative to the
+ * current pc.
+ */
+void x86_call( struct x86_function *p, void (*label)())
+{
+ emit_1ub(p, 0xe8);
+ emit_1i(p, cptr(label) - x86_get_label(p) - 4);
+}
+#else
+void x86_call( struct x86_function *p, struct x86_reg reg)
+{
+ emit_1ub(p, 0xff);
+ emit_modrm(p, reg, reg);
+}
+#endif
+
+
+/* michal:
+ * Temporary. As I need immediate operands, and dont want to mess with the codegen,
+ * I load the immediate into general purpose register and use it.
+ */
+void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+ assert(dst.mod == mod_REG);
+ emit_1ub(p, 0xb8 + dst.idx);
+ emit_1i(p, imm);
+}
+
+void x86_push( struct x86_function *p,
+ struct x86_reg reg )
+{
+ assert(reg.mod == mod_REG);
+ emit_1ub(p, 0x50 + reg.idx);
+ p->stack_offset += 4;
+}
+
+void x86_pop( struct x86_function *p,
+ struct x86_reg reg )
+{
+ assert(reg.mod == mod_REG);
+ emit_1ub(p, 0x58 + reg.idx);
+ p->stack_offset -= 4;
+}
+
+void x86_inc( struct x86_function *p,
+ struct x86_reg reg )
+{
+ assert(reg.mod == mod_REG);
+ emit_1ub(p, 0x40 + reg.idx);
+}
+
+void x86_dec( struct x86_function *p,
+ struct x86_reg reg )
+{
+ assert(reg.mod == mod_REG);
+ emit_1ub(p, 0x48 + reg.idx);
+}
+
+void x86_ret( struct x86_function *p )
+{
+ emit_1ub(p, 0xc3);
+}
+
+void x86_sahf( struct x86_function *p )
+{
+ emit_1ub(p, 0x9e);
+}
+
+void x86_mov( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_op_modrm( p, 0x8b, 0x89, dst, src );
+}
+
+void x86_xor( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_op_modrm( p, 0x33, 0x31, dst, src );
+}
+
+void x86_cmp( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_op_modrm( p, 0x3b, 0x39, dst, src );
+}
+
+void x86_lea( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_1ub(p, 0x8d);
+ emit_modrm( p, dst, src );
+}
+
+void x86_test( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_1ub(p, 0x85);
+ emit_modrm( p, dst, src );
+}
+
+void x86_add( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_op_modrm(p, 0x03, 0x01, dst, src );
+}
+
+void x86_mul( struct x86_function *p,
+ struct x86_reg src )
+{
+ assert (src.file == file_REG32 && src.mod == mod_REG);
+ emit_op_modrm(p, 0xf7, 0, x86_make_reg (file_REG32, reg_SP), src );
+}
+
+void x86_sub( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_op_modrm(p, 0x2b, 0x29, dst, src );
+}
+
+void x86_or( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_op_modrm( p, 0x0b, 0x09, dst, src );
+}
+
+void x86_and( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_op_modrm( p, 0x23, 0x21, dst, src );
+}
+
+
+
+/***********************************************************************
+ * SSE instructions
+ */
+
+
+void sse_movss( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, 0xF3, X86_TWOB);
+ emit_op_modrm( p, 0x10, 0x11, dst, src );
+}
+
+void sse_movaps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_1ub(p, X86_TWOB);
+ emit_op_modrm( p, 0x28, 0x29, dst, src );
+}
+
+void sse_movups( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_1ub(p, X86_TWOB);
+ emit_op_modrm( p, 0x10, 0x11, dst, src );
+}
+
+void sse_movhps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ assert(dst.mod != mod_REG || src.mod != mod_REG);
+ emit_1ub(p, X86_TWOB);
+ emit_op_modrm( p, 0x16, 0x17, dst, src ); /* cf movlhps */
+}
+
+void sse_movlps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ assert(dst.mod != mod_REG || src.mod != mod_REG);
+ emit_1ub(p, X86_TWOB);
+ emit_op_modrm( p, 0x12, 0x13, dst, src ); /* cf movhlps */
+}
+
+void sse_maxps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x5F);
+ emit_modrm( p, dst, src );
+}
+
+void sse_maxss( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0xF3, X86_TWOB, 0x5F);
+ emit_modrm( p, dst, src );
+}
+
+void sse_divss( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0xF3, X86_TWOB, 0x5E);
+ emit_modrm( p, dst, src );
+}
+
+void sse_minps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x5D);
+ emit_modrm( p, dst, src );
+}
+
+void sse_subps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x5C);
+ emit_modrm( p, dst, src );
+}
+
+void sse_mulps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x59);
+ emit_modrm( p, dst, src );
+}
+
+void sse_mulss( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0xF3, X86_TWOB, 0x59);
+ emit_modrm( p, dst, src );
+}
+
+void sse_addps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x58);
+ emit_modrm( p, dst, src );
+}
+
+void sse_addss( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0xF3, X86_TWOB, 0x58);
+ emit_modrm( p, dst, src );
+}
+
+void sse_andnps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x55);
+ emit_modrm( p, dst, src );
+}
+
+void sse_andps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x54);
+ emit_modrm( p, dst, src );
+}
+
+void sse_rsqrtps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x52);
+ emit_modrm( p, dst, src );
+}
+
+void sse_rsqrtss( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0xF3, X86_TWOB, 0x52);
+ emit_modrm( p, dst, src );
+
+}
+
+void sse_movhlps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ assert(dst.mod == mod_REG && src.mod == mod_REG);
+ emit_2ub(p, X86_TWOB, 0x12);
+ emit_modrm( p, dst, src );
+}
+
+void sse_movlhps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ assert(dst.mod == mod_REG && src.mod == mod_REG);
+ emit_2ub(p, X86_TWOB, 0x16);
+ emit_modrm( p, dst, src );
+}
+
+void sse_orps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x56);
+ emit_modrm( p, dst, src );
+}
+
+void sse_xorps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x57);
+ emit_modrm( p, dst, src );
+}
+
+void sse_cvtps2pi( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ assert(dst.file == file_MMX &&
+ (src.file == file_XMM || src.mod != mod_REG));
+
+ p->need_emms = 1;
+
+ emit_2ub(p, X86_TWOB, 0x2d);
+ emit_modrm( p, dst, src );
+}
+
+
+/* Shufps can also be used to implement a reduced swizzle when dest ==
+ * arg0.
+ */
+void sse_shufps( struct x86_function *p,
+ struct x86_reg dest,
+ struct x86_reg arg0,
+ unsigned char shuf)
+{
+ emit_2ub(p, X86_TWOB, 0xC6);
+ emit_modrm(p, dest, arg0);
+ emit_1ub(p, shuf);
+}
+
+void sse_cmpps( struct x86_function *p,
+ struct x86_reg dest,
+ struct x86_reg arg0,
+ unsigned char cc)
+{
+ emit_2ub(p, X86_TWOB, 0xC2);
+ emit_modrm(p, dest, arg0);
+ emit_1ub(p, cc);
+}
+
+void sse_pmovmskb( struct x86_function *p,
+ struct x86_reg dest,
+ struct x86_reg src)
+{
+ emit_3ub(p, 0x66, X86_TWOB, 0xD7);
+ emit_modrm(p, dest, src);
+}
+
+/***********************************************************************
+ * SSE2 instructions
+ */
+
+/**
+ * Perform a reduced swizzle:
+ */
+void sse2_pshufd( struct x86_function *p,
+ struct x86_reg dest,
+ struct x86_reg arg0,
+ unsigned char shuf)
+{
+ emit_3ub(p, 0x66, X86_TWOB, 0x70);
+ emit_modrm(p, dest, arg0);
+ emit_1ub(p, shuf);
+}
+
+void sse2_cvttps2dq( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub( p, 0xF3, X86_TWOB, 0x5B );
+ emit_modrm( p, dst, src );
+}
+
+void sse2_cvtps2dq( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0x66, X86_TWOB, 0x5B);
+ emit_modrm( p, dst, src );
+}
+
+void sse2_packssdw( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0x66, X86_TWOB, 0x6B);
+ emit_modrm( p, dst, src );
+}
+
+void sse2_packsswb( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0x66, X86_TWOB, 0x63);
+ emit_modrm( p, dst, src );
+}
+
+void sse2_packuswb( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0x66, X86_TWOB, 0x67);
+ emit_modrm( p, dst, src );
+}
+
+void sse2_rcpps( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, X86_TWOB, 0x53);
+ emit_modrm( p, dst, src );
+}
+
+void sse2_rcpss( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_3ub(p, 0xF3, X86_TWOB, 0x53);
+ emit_modrm( p, dst, src );
+}
+
+void sse2_movd( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ emit_2ub(p, 0x66, X86_TWOB);
+ emit_op_modrm( p, 0x6e, 0x7e, dst, src );
+}
+
+
+
+
+/***********************************************************************
+ * x87 instructions
+ */
+void x87_fist( struct x86_function *p, struct x86_reg dst )
+{
+ emit_1ub(p, 0xdb);
+ emit_modrm_noreg(p, 2, dst);
+}
+
+void x87_fistp( struct x86_function *p, struct x86_reg dst )
+{
+ emit_1ub(p, 0xdb);
+ emit_modrm_noreg(p, 3, dst);
+}
+
+void x87_fild( struct x86_function *p, struct x86_reg arg )
+{
+ emit_1ub(p, 0xdf);
+ emit_modrm_noreg(p, 0, arg);
+}
+
+void x87_fldz( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xee);
+}
+
+
+void x87_fldcw( struct x86_function *p, struct x86_reg arg )
+{
+ assert(arg.file == file_REG32);
+ assert(arg.mod != mod_REG);
+ emit_1ub(p, 0xd9);
+ emit_modrm_noreg(p, 5, arg);
+}
+
+void x87_fld1( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xe8);
+}
+
+void x87_fldl2e( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xea);
+}
+
+void x87_fldln2( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xed);
+}
+
+void x87_fwait( struct x86_function *p )
+{
+ emit_1ub(p, 0x9b);
+}
+
+void x87_fnclex( struct x86_function *p )
+{
+ emit_2ub(p, 0xdb, 0xe2);
+}
+
+void x87_fclex( struct x86_function *p )
+{
+ x87_fwait(p);
+ x87_fnclex(p);
+}
+
+
+static void x87_arith_op( struct x86_function *p, struct x86_reg dst, struct x86_reg arg,
+ unsigned char dst0ub0,
+ unsigned char dst0ub1,
+ unsigned char arg0ub0,
+ unsigned char arg0ub1,
+ unsigned char argmem_noreg)
+{
+ assert(dst.file == file_x87);
+
+ if (arg.file == file_x87) {
+ if (dst.idx == 0)
+ emit_2ub(p, dst0ub0, dst0ub1+arg.idx);
+ else if (arg.idx == 0)
+ emit_2ub(p, arg0ub0, arg0ub1+arg.idx);
+ else
+ assert(0);
+ }
+ else if (dst.idx == 0) {
+ assert(arg.file == file_REG32);
+ emit_1ub(p, 0xd8);
+ emit_modrm_noreg(p, argmem_noreg, arg);
+ }
+ else
+ assert(0);
+}
+
+void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+ x87_arith_op(p, dst, arg,
+ 0xd8, 0xc8,
+ 0xdc, 0xc8,
+ 4);
+}
+
+void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+ x87_arith_op(p, dst, arg,
+ 0xd8, 0xe0,
+ 0xdc, 0xe8,
+ 4);
+}
+
+void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+ x87_arith_op(p, dst, arg,
+ 0xd8, 0xe8,
+ 0xdc, 0xe0,
+ 5);
+}
+
+void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+ x87_arith_op(p, dst, arg,
+ 0xd8, 0xc0,
+ 0xdc, 0xc0,
+ 0);
+}
+
+void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+ x87_arith_op(p, dst, arg,
+ 0xd8, 0xf0,
+ 0xdc, 0xf8,
+ 6);
+}
+
+void x87_fdivr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+ x87_arith_op(p, dst, arg,
+ 0xd8, 0xf8,
+ 0xdc, 0xf0,
+ 7);
+}
+
+void x87_fmulp( struct x86_function *p, struct x86_reg dst )
+{
+ assert(dst.file == file_x87);
+ assert(dst.idx >= 1);
+ emit_2ub(p, 0xde, 0xc8+dst.idx);
+}
+
+void x87_fsubp( struct x86_function *p, struct x86_reg dst )
+{
+ assert(dst.file == file_x87);
+ assert(dst.idx >= 1);
+ emit_2ub(p, 0xde, 0xe8+dst.idx);
+}
+
+void x87_fsubrp( struct x86_function *p, struct x86_reg dst )
+{
+ assert(dst.file == file_x87);
+ assert(dst.idx >= 1);
+ emit_2ub(p, 0xde, 0xe0+dst.idx);
+}
+
+void x87_faddp( struct x86_function *p, struct x86_reg dst )
+{
+ assert(dst.file == file_x87);
+ assert(dst.idx >= 1);
+ emit_2ub(p, 0xde, 0xc0+dst.idx);
+}
+
+void x87_fdivp( struct x86_function *p, struct x86_reg dst )
+{
+ assert(dst.file == file_x87);
+ assert(dst.idx >= 1);
+ emit_2ub(p, 0xde, 0xf8+dst.idx);
+}
+
+void x87_fdivrp( struct x86_function *p, struct x86_reg dst )
+{
+ assert(dst.file == file_x87);
+ assert(dst.idx >= 1);
+ emit_2ub(p, 0xde, 0xf0+dst.idx);
+}
+
+void x87_fucom( struct x86_function *p, struct x86_reg arg )
+{
+ assert(arg.file == file_x87);
+ emit_2ub(p, 0xdd, 0xe0+arg.idx);
+}
+
+void x87_fucomp( struct x86_function *p, struct x86_reg arg )
+{
+ assert(arg.file == file_x87);
+ emit_2ub(p, 0xdd, 0xe8+arg.idx);
+}
+
+void x87_fucompp( struct x86_function *p )
+{
+ emit_2ub(p, 0xda, 0xe9);
+}
+
+void x87_fxch( struct x86_function *p, struct x86_reg arg )
+{
+ assert(arg.file == file_x87);
+ emit_2ub(p, 0xd9, 0xc8+arg.idx);
+}
+
+void x87_fabs( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xe1);
+}
+
+void x87_fchs( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xe0);
+}
+
+void x87_fcos( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xff);
+}
+
+
+void x87_fprndint( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xfc);
+}
+
+void x87_fscale( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xfd);
+}
+
+void x87_fsin( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xfe);
+}
+
+void x87_fsincos( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xfb);
+}
+
+void x87_fsqrt( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xfa);
+}
+
+void x87_fxtract( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xf4);
+}
+
+/* st0 = (2^st0)-1
+ *
+ * Restrictions: -1.0 <= st0 <= 1.0
+ */
+void x87_f2xm1( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xf0);
+}
+
+/* st1 = st1 * log2(st0);
+ * pop_stack;
+ */
+void x87_fyl2x( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xf1);
+}
+
+/* st1 = st1 * log2(st0 + 1.0);
+ * pop_stack;
+ *
+ * A fast operation, with restrictions: -.29 < st0 < .29
+ */
+void x87_fyl2xp1( struct x86_function *p )
+{
+ emit_2ub(p, 0xd9, 0xf9);
+}
+
+
+void x87_fld( struct x86_function *p, struct x86_reg arg )
+{
+ if (arg.file == file_x87)
+ emit_2ub(p, 0xd9, 0xc0 + arg.idx);
+ else {
+ emit_1ub(p, 0xd9);
+ emit_modrm_noreg(p, 0, arg);
+ }
+}
+
+void x87_fst( struct x86_function *p, struct x86_reg dst )
+{
+ if (dst.file == file_x87)
+ emit_2ub(p, 0xdd, 0xd0 + dst.idx);
+ else {
+ emit_1ub(p, 0xd9);
+ emit_modrm_noreg(p, 2, dst);
+ }
+}
+
+void x87_fstp( struct x86_function *p, struct x86_reg dst )
+{
+ if (dst.file == file_x87)
+ emit_2ub(p, 0xdd, 0xd8 + dst.idx);
+ else {
+ emit_1ub(p, 0xd9);
+ emit_modrm_noreg(p, 3, dst);
+ }
+}
+
+void x87_fcom( struct x86_function *p, struct x86_reg dst )
+{
+ if (dst.file == file_x87)
+ emit_2ub(p, 0xd8, 0xd0 + dst.idx);
+ else {
+ emit_1ub(p, 0xd8);
+ emit_modrm_noreg(p, 2, dst);
+ }
+}
+
+void x87_fcomp( struct x86_function *p, struct x86_reg dst )
+{
+ if (dst.file == file_x87)
+ emit_2ub(p, 0xd8, 0xd8 + dst.idx);
+ else {
+ emit_1ub(p, 0xd8);
+ emit_modrm_noreg(p, 3, dst);
+ }
+}
+
+
+void x87_fnstsw( struct x86_function *p, struct x86_reg dst )
+{
+ assert(dst.file == file_REG32);
+
+ if (dst.idx == reg_AX &&
+ dst.mod == mod_REG)
+ emit_2ub(p, 0xdf, 0xe0);
+ else {
+ emit_1ub(p, 0xdd);
+ emit_modrm_noreg(p, 7, dst);
+ }
+}
+
+
+
+
+/***********************************************************************
+ * MMX instructions
+ */
+
+void mmx_emms( struct x86_function *p )
+{
+ assert(p->need_emms);
+ emit_2ub(p, 0x0f, 0x77);
+ p->need_emms = 0;
+}
+
+void mmx_packssdw( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ assert(dst.file == file_MMX &&
+ (src.file == file_MMX || src.mod != mod_REG));
+
+ p->need_emms = 1;
+
+ emit_2ub(p, X86_TWOB, 0x6b);
+ emit_modrm( p, dst, src );
+}
+
+void mmx_packuswb( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ assert(dst.file == file_MMX &&
+ (src.file == file_MMX || src.mod != mod_REG));
+
+ p->need_emms = 1;
+
+ emit_2ub(p, X86_TWOB, 0x67);
+ emit_modrm( p, dst, src );
+}
+
+void mmx_movd( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ p->need_emms = 1;
+ emit_1ub(p, X86_TWOB);
+ emit_op_modrm( p, 0x6e, 0x7e, dst, src );
+}
+
+void mmx_movq( struct x86_function *p,
+ struct x86_reg dst,
+ struct x86_reg src )
+{
+ p->need_emms = 1;
+ emit_1ub(p, X86_TWOB);
+ emit_op_modrm( p, 0x6f, 0x7f, dst, src );
+}
+
+
+/***********************************************************************
+ * Helper functions
+ */
+
+
+/* Retreive a reference to one of the function arguments, taking into
+ * account any push/pop activity:
+ */
+struct x86_reg x86_fn_arg( struct x86_function *p,
+ unsigned arg )
+{
+ return x86_make_disp(x86_make_reg(file_REG32, reg_SP),
+ p->stack_offset + arg * 4); /* ??? */
+}
+
+
+void x86_init_func( struct x86_function *p )
+{
+ p->size = 0;
+ p->store = NULL;
+ p->csr = p->store;
+}
+
+void x86_init_func_size( struct x86_function *p, unsigned code_size )
+{
+ p->size = code_size;
+ p->store = _mesa_exec_malloc(code_size);
+ p->csr = p->store;
+}
+
+void x86_release_func( struct x86_function *p )
+{
+ _mesa_exec_free(p->store);
+ p->store = NULL;
+ p->csr = NULL;
+ p->size = 0;
+}
+
+
+void (*x86_get_func( struct x86_function *p ))(void)
+{
+ if (DISASSEM && p->store)
+ _mesa_printf("disassemble %p %p\n", p->store, p->csr);
+ return (void (*)(void)) (unsigned long) p->store;
+}
+
+#else
+
+void x86sse_dummy( void )
+{
+}
+
+#endif
diff --git a/src/gallium/auxiliary/rtasm/x86sse.h b/src/gallium/auxiliary/rtasm/x86sse.h
new file mode 100644
index 0000000000..c2aa416492
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/x86sse.h
@@ -0,0 +1,256 @@
+
+#ifndef _X86SSE_H_
+#define _X86SSE_H_
+
+#if defined(__i386__) || defined(__386__)
+
+/* It is up to the caller to ensure that instructions issued are
+ * suitable for the host cpu. There are no checks made in this module
+ * for mmx/sse/sse2 support on the cpu.
+ */
+struct x86_reg {
+ unsigned file:3;
+ unsigned idx:3;
+ unsigned mod:2; /* mod_REG if this is just a register */
+ int disp:24; /* only +/- 23bits of offset - should be enough... */
+};
+
+struct x86_function {
+ unsigned size;
+ unsigned char *store;
+ unsigned char *csr;
+ unsigned stack_offset;
+ int need_emms;
+ const char *fn;
+};
+
+enum x86_reg_file {
+ file_REG32,
+ file_MMX,
+ file_XMM,
+ file_x87
+};
+
+/* Values for mod field of modr/m byte
+ */
+enum x86_reg_mod {
+ mod_INDIRECT,
+ mod_DISP8,
+ mod_DISP32,
+ mod_REG
+};
+
+enum x86_reg_name {
+ reg_AX,
+ reg_CX,
+ reg_DX,
+ reg_BX,
+ reg_SP,
+ reg_BP,
+ reg_SI,
+ reg_DI
+};
+
+
+enum x86_cc {
+ cc_O, /* overflow */
+ cc_NO, /* not overflow */
+ cc_NAE, /* not above or equal / carry */
+ cc_AE, /* above or equal / not carry */
+ cc_E, /* equal / zero */
+ cc_NE /* not equal / not zero */
+};
+
+enum sse_cc {
+ cc_Equal,
+ cc_LessThan,
+ cc_LessThanEqual,
+ cc_Unordered,
+ cc_NotEqual,
+ cc_NotLessThan,
+ cc_NotLessThanEqual,
+ cc_Ordered
+};
+
+#define cc_Z cc_E
+#define cc_NZ cc_NE
+
+/* Begin/end/retreive function creation:
+ */
+
+
+void x86_init_func( struct x86_function *p );
+void x86_init_func_size( struct x86_function *p, unsigned code_size );
+void x86_release_func( struct x86_function *p );
+void (*x86_get_func( struct x86_function *p ))( void );
+
+
+
+/* Create and manipulate registers and regmem values:
+ */
+struct x86_reg x86_make_reg( enum x86_reg_file file,
+ enum x86_reg_name idx );
+
+struct x86_reg x86_make_disp( struct x86_reg reg,
+ int disp );
+
+struct x86_reg x86_deref( struct x86_reg reg );
+
+struct x86_reg x86_get_base_reg( struct x86_reg reg );
+
+
+/* Labels, jumps and fixup:
+ */
+unsigned char *x86_get_label( struct x86_function *p );
+
+void x86_jcc( struct x86_function *p,
+ enum x86_cc cc,
+ unsigned char *label );
+
+unsigned char *x86_jcc_forward( struct x86_function *p,
+ enum x86_cc cc );
+
+unsigned char *x86_jmp_forward( struct x86_function *p);
+
+unsigned char *x86_call_forward( struct x86_function *p);
+
+void x86_fixup_fwd_jump( struct x86_function *p,
+ unsigned char *fixup );
+
+void x86_jmp( struct x86_function *p, unsigned char *label );
+
+/* void x86_call( struct x86_function *p, void (*label)() ); */
+void x86_call( struct x86_function *p, struct x86_reg reg);
+
+/* michal:
+ * Temporary. As I need immediate operands, and dont want to mess with the codegen,
+ * I load the immediate into general purpose register and use it.
+ */
+void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm );
+
+
+/* Macro for sse_shufps() and sse2_pshufd():
+ */
+#define SHUF(_x,_y,_z,_w) (((_x)<<0) | ((_y)<<2) | ((_z)<<4) | ((_w)<<6))
+#define SHUF_NOOP RSW(0,1,2,3)
+#define GET_SHUF(swz, idx) (((swz) >> ((idx)*2)) & 0x3)
+
+void mmx_emms( struct x86_function *p );
+void mmx_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_movq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+
+void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_cvttps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packsswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
+ unsigned char shuf );
+void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+
+void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_divss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_andnps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src,
+ unsigned char cc );
+void sse_maxps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_maxss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_minps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movaps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movhlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movlhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movups( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_mulss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_orps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_xorps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_subps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_rsqrtps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
+ unsigned char shuf );
+void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src );
+
+void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_cmp( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_dec( struct x86_function *p, struct x86_reg reg );
+void x86_inc( struct x86_function *p, struct x86_reg reg );
+void x86_lea( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_mov( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_mul( struct x86_function *p, struct x86_reg src );
+void x86_or( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_pop( struct x86_function *p, struct x86_reg reg );
+void x86_push( struct x86_function *p, struct x86_reg reg );
+void x86_ret( struct x86_function *p );
+void x86_sub( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_sahf( struct x86_function *p );
+
+void x87_f2xm1( struct x86_function *p );
+void x87_fabs( struct x86_function *p );
+void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_faddp( struct x86_function *p, struct x86_reg dst );
+void x87_fchs( struct x86_function *p );
+void x87_fclex( struct x86_function *p );
+void x87_fcom( struct x86_function *p, struct x86_reg dst );
+void x87_fcomp( struct x86_function *p, struct x86_reg dst );
+void x87_fcos( struct x86_function *p );
+void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fdivp( struct x86_function *p, struct x86_reg dst );
+void x87_fdivr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fdivrp( struct x86_function *p, struct x86_reg dst );
+void x87_fild( struct x86_function *p, struct x86_reg arg );
+void x87_fist( struct x86_function *p, struct x86_reg dst );
+void x87_fistp( struct x86_function *p, struct x86_reg dst );
+void x87_fld( struct x86_function *p, struct x86_reg arg );
+void x87_fld1( struct x86_function *p );
+void x87_fldcw( struct x86_function *p, struct x86_reg arg );
+void x87_fldl2e( struct x86_function *p );
+void x87_fldln2( struct x86_function *p );
+void x87_fldz( struct x86_function *p );
+void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fmulp( struct x86_function *p, struct x86_reg dst );
+void x87_fnclex( struct x86_function *p );
+void x87_fprndint( struct x86_function *p );
+void x87_fscale( struct x86_function *p );
+void x87_fsin( struct x86_function *p );
+void x87_fsincos( struct x86_function *p );
+void x87_fsqrt( struct x86_function *p );
+void x87_fst( struct x86_function *p, struct x86_reg dst );
+void x87_fstp( struct x86_function *p, struct x86_reg dst );
+void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fsubp( struct x86_function *p, struct x86_reg dst );
+void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fsubrp( struct x86_function *p, struct x86_reg dst );
+void x87_fxch( struct x86_function *p, struct x86_reg dst );
+void x87_fxtract( struct x86_function *p );
+void x87_fyl2x( struct x86_function *p );
+void x87_fyl2xp1( struct x86_function *p );
+void x87_fwait( struct x86_function *p );
+void x87_fnstsw( struct x86_function *p, struct x86_reg dst );
+void x87_fucompp( struct x86_function *p );
+void x87_fucomp( struct x86_function *p, struct x86_reg arg );
+void x87_fucom( struct x86_function *p, struct x86_reg arg );
+
+
+
+/* Retreive a reference to one of the function arguments, taking into
+ * account any push/pop activity. Note - doesn't track explict
+ * manipulation of ESP by other instructions.
+ */
+struct x86_reg x86_fn_arg( struct x86_function *p, unsigned arg );
+
+#endif
+#endif