From 9ac9605de156408580b81ba7e2780bd3f5372c6d Mon Sep 17 00:00:00 2001
From: Michal Krol <mjkrol@gmail.org>
Date: Mon, 27 Feb 2006 14:41:41 +0000
Subject: More GLSL code: - add x86 code generator; - add full support for
 uniforms in ARB_shader_objects; - add assembly instruction: global_addr; -
 reorganize #includes; - built-in uniforms accessed by index, rather than by
 name; - add some entries to x86sse rtasm; - add configurations to VC6
 projects: 'Release x86' and 'Debug x86'; - #define SLANG_X86 active only on
 VC6 x86 builds; - introduce code export table for a shader; - remove GNU
 license from the noise library;

---
 src/mesa/x86/rtasm/x86sse.c | 83 +++++++++++++++++++++++++++++++++++++++------
 src/mesa/x86/rtasm/x86sse.h | 32 +++++++++++++----
 2 files changed, 98 insertions(+), 17 deletions(-)

(limited to 'src/mesa/x86')

diff --git a/src/mesa/x86/rtasm/x86sse.c b/src/mesa/x86/rtasm/x86sse.c
index 0c9ffe25fa..82a18f012a 100644
--- a/src/mesa/x86/rtasm/x86sse.c
+++ b/src/mesa/x86/rtasm/x86sse.c
@@ -1,4 +1,4 @@
-#if defined(USE_X86_ASM)
+#if defined(USE_X86_ASM) || defined(SLANG_X86)
 
 #include "imports.h"
 #include "x86sse.h"
@@ -85,10 +85,10 @@ static void emit_modrm( struct x86_function *p,
    case mod_INDIRECT:
       break;
    case mod_DISP8:
-      emit_1b(p, regmem.disp);
+      emit_1b(p, regmem.disp);
       break;
    case mod_DISP32:
-      emit_1i(p, regmem.disp);
+      emit_1i(p, regmem.disp);
       break;
    default:
       assert(0);
@@ -142,8 +142,8 @@ static void emit_op_modrm( struct x86_function *p,
 
 /* Create and manipulate registers and regmem values:
  */
-struct x86_reg x86_make_reg( GLuint file,
-			     GLuint idx )
+struct x86_reg x86_make_reg( enum x86_reg_file file,
+			     enum x86_reg_name idx )
 {
    struct x86_reg reg;
 
@@ -198,7 +198,7 @@ GLubyte *x86_get_label( struct x86_function *p )
 
 
 void x86_jcc( struct x86_function *p,
-	      GLuint cc,
+	      enum x86_cc cc,
 	      GLubyte *label )
 {
    GLint offset = label - (x86_get_label(p) + 2);
@@ -217,11 +217,25 @@ void x86_jcc( struct x86_function *p,
 /* Always use a 32bit offset for forward jumps:
  */
 GLubyte *x86_jcc_forward( struct x86_function *p,
-			  GLuint cc )
+			  enum x86_cc cc )
 {
    emit_2ub(p, 0x0f, 0x80 + cc);
    emit_1i(p, 0);
    return x86_get_label(p);
+}
+
+GLubyte *x86_jmp_forward( struct x86_function *p)
+{
+   emit_1ub(p, 0xe9);
+   emit_1i(p, 0);
+   return x86_get_label(p);
+}
+
+GLubyte *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:
@@ -230,6 +244,29 @@ void x86_fixup_fwd_jump( struct x86_function *p,
 			 GLubyte *fixup )
 {
    *(int *)(fixup - 4) = x86_get_label(p) - fixup;
+}
+
+void x86_jmp( struct x86_function *p, GLubyte *label)
+{
+   emit_1ub(p, 0xe9);
+   emit_1i(p, label - x86_get_label(p) - 4);
+}
+
+void x86_call( struct x86_function *p, GLubyte *label)
+{
+   emit_1ub(p, 0xe8);
+   emit_1i(p, label - x86_get_label(p) - 4);
+}
+
+/* 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, GLint imm )
+{
+   assert(dst.mod == mod_REG);
+   emit_1ub(p, 0xb8 + dst.idx);
+   emit_1i(p, imm);
 }
 
 void x86_push( struct x86_function *p,
@@ -307,6 +344,27 @@ void x86_test( struct x86_function *p,
 {
    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 );
 }
 
 
@@ -971,8 +1029,13 @@ struct x86_reg x86_fn_arg( struct x86_function *p,
 
 void x86_init_func( struct x86_function *p )
 {
-   p->store = _mesa_exec_malloc(1024);
-   p->csr = p->store;
+   x86_init_func_size(p, 1024);
+}
+
+void x86_init_func_size( struct x86_function *p, GLuint code_size )
+{
+   p->store = _mesa_exec_malloc(code_size);
+   p->csr = p->store;
 }
 
 void x86_release_func( struct x86_function *p )
@@ -985,7 +1048,7 @@ void (*x86_get_func( struct x86_function *p ))(void)
 {
    if (DISASSEM)
       _mesa_printf("disassemble %p %p\n", p->store, p->csr);
-   return (void (*)())p->store;
+   return (void (*)(void))p->store;
 }
 
 #else
diff --git a/src/mesa/x86/rtasm/x86sse.h b/src/mesa/x86/rtasm/x86sse.h
index 611d01e1cb..55a98564af 100644
--- a/src/mesa/x86/rtasm/x86sse.h
+++ b/src/mesa/x86/rtasm/x86sse.h
@@ -2,7 +2,7 @@
 #ifndef _X86SSE_H_
 #define _X86SSE_H_
 
-#if defined(USE_X86_ASM)
+#if defined(USE_X86_ASM) || defined(SLANG_X86)
 
 #include "glheader.h"
 
@@ -80,7 +80,8 @@ enum sse_cc {
  */
 
 
-void x86_init_func( struct x86_function *p );
+void x86_init_func( struct x86_function *p );
+void x86_init_func_size( struct x86_function *p, GLuint code_size );
 void x86_release_func( struct x86_function *p );
 void (*x86_get_func( struct x86_function *p ))( void );
 
@@ -108,10 +109,24 @@ void x86_jcc( struct x86_function *p,
 	      GLubyte *label );
 
 GLubyte *x86_jcc_forward( struct x86_function *p,
-			  enum x86_cc cc );
+			  enum x86_cc cc );
+
+GLubyte *x86_jmp_forward( struct x86_function *p);
+
+GLubyte *x86_call_forward( struct x86_function *p);
 
 void x86_fixup_fwd_jump( struct x86_function *p,
-			 GLubyte *fixup );
+			 GLubyte *fixup );
+
+void x86_jmp( struct x86_function *p, GLubyte *label );
+
+void x86_call( struct x86_function *p, GLubyte *label );
+
+/* 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, GLint imm );
 
 
 /* Macro for sse_shufps() and sse2_pshufd():
@@ -153,15 +168,18 @@ void sse_mulps( 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_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, GLubyte shuf );
-
+
+void x86_add( 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_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_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_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 );
-- 
cgit v1.2.3