Merge branch 'mesa_7_5_branch'

[mesa.git] / src / gallium / auxiliary / rtasm / rtasm_x86sse.c

diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
index d78676b8f36a6d8b206519ae91f1f31d9e749c8a..1acf3c373ebee1700b74463c834d1363a2098e3d 100644 (file)
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -21,11 +21,13 @@
  *
  **************************************************************************/
 
-#if defined(__i386__) || defined(__386__) || defined(i386)
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_X86)
 
 #include "pipe/p_compiler.h"
-#include "pipe/p_debug.h"
-#include "pipe/p_pointer.h"
+#include "util/u_debug.h"
+#include "util/u_pointer.h"
 
 #include "rtasm_execmem.h"
 #include "rtasm_x86sse.h"
@@ -218,6 +220,8 @@ static void emit_3ub( struct x86_function *p, unsigned char b0, unsigned char b1
 
 /* Build a modRM byte + possible displacement.  No treatment of SIB
  * indexing.  BZZT - no way to encode an absolute address.
+ *
+ * This is the "/r" field in the x86 manuals...
  */
 static void emit_modrm( struct x86_function *p, 
                        struct x86_reg reg, 
@@ -236,7 +240,8 @@ static void emit_modrm( struct x86_function *p,
    /* Oh-oh we've stumbled into the SIB thing.
     */
    if (regmem.file == file_REG32 &&
-       regmem.idx == reg_SP) {
+       regmem.idx == reg_SP &&
+       regmem.mod != mod_REG) {
       emit_1ub(p, 0x24);               /* simplistic! */
    }
 
@@ -256,7 +261,8 @@ static void emit_modrm( struct x86_function *p,
    }
 }
 
-
+/* Emits the "/0".."/7" specialized versions of the modrm ("/r") bytes.
+ */
 static void emit_modrm_noreg( struct x86_function *p,
                              unsigned op,
                              struct x86_reg regmem )
@@ -325,7 +331,7 @@ struct x86_reg x86_make_disp( struct x86_reg reg,
    else
       reg.disp += disp;
 
-   if (reg.disp == 0)
+   if (reg.disp == 0 && reg.idx != reg_BP)
       reg.mod = mod_INDIRECT;
    else if (reg.disp <= 127 && reg.disp >= -128)
       reg.mod = mod_DISP8;
@@ -365,8 +371,11 @@ void x86_jcc( struct x86_function *p,
    DUMP_I(cc);
    
    if (offset < 0) {
-      int amt = p->csr - p->store;
-      assert(amt > -offset);
+      /*assert(p->csr - p->store > -offset);*/
+      if (p->csr - p->store <= -offset) {
+         /* probably out of memory (using the error_overflow buffer) */
+         return;
+      }
    }
 
    if (offset <= 127 && offset >= -128) {
@@ -431,18 +440,73 @@ 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 )
 {
    DUMP_RI( dst, imm );
+   assert(dst.file == file_REG32);
    assert(dst.mod == mod_REG);
    emit_1ub(p, 0xb8 + dst.idx);
    emit_1i(p, imm);
 }
 
+/**
+ * Immediate group 1 instructions.
+ */
+static INLINE void 
+x86_group1_imm( struct x86_function *p, 
+                unsigned op, struct x86_reg dst, int imm )
+{
+   assert(dst.file == file_REG32);
+   assert(dst.mod == mod_REG);
+   if(-0x80 <= imm && imm < 0x80) {
+      emit_1ub(p, 0x83);
+      emit_modrm_noreg(p, op, dst);
+      emit_1b(p, (char)imm);
+   }
+   else {
+      emit_1ub(p, 0x81);
+      emit_modrm_noreg(p, op, dst);
+      emit_1i(p, imm);
+   }
+}
+
+void x86_add_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 0, dst, imm);
+}
+
+void x86_or_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 1, dst, imm);
+}
+
+void x86_and_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 4, dst, imm);
+}
+
+void x86_sub_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 5, dst, imm);
+}
+
+void x86_xor_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 6, dst, imm);
+}
+
+void x86_cmp_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 7, dst, imm);
+}
+
+
 void x86_push( struct x86_function *p,
               struct x86_reg reg )
 {
@@ -459,6 +523,17 @@ void x86_push( struct x86_function *p,
    p->stack_offset += 4;
 }
 
+void x86_push_imm32( struct x86_function *p,
+                     int imm32 )
+{
+   DUMP_I( imm32 );
+   emit_1ub(p, 0x68);
+   emit_1i(p,  imm32);
+
+   p->stack_offset += 4;
+}
+
+
 void x86_pop( struct x86_function *p,
              struct x86_reg reg )
 {
@@ -604,6 +679,44 @@ void x86_and( struct x86_function *p,
  * SSE instructions
  */
 
+void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 0, ptr);
+}
+
+void sse_prefetch0( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 1, ptr);
+}
+
+void sse_prefetch1( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 2, ptr);
+}
+
+void sse_movntps( struct x86_function *p, 
+                  struct x86_reg dst,
+                  struct x86_reg src)
+{
+   DUMP_RR( dst, src );
+
+   assert(dst.mod != mod_REG);
+   assert(src.mod == mod_REG);
+   emit_2ub(p, 0x0f, 0x2b);
+   emit_modrm(p, src, dst);
+}
+
+
+
 
 void sse_movss( struct x86_function *p,
                struct x86_reg dst,
@@ -863,7 +976,7 @@ void sse_unpcklps( struct x86_function *p, struct x86_reg dst, struct x86_reg sr
 void sse_cmpps( struct x86_function *p,
                struct x86_reg dst,
                struct x86_reg src,
-               unsigned char cc) 
+               enum sse_cc cc) 
 {
    DUMP_RRI( dst, src, cc );
    emit_2ub(p, X86_TWOB, 0xC2);
@@ -880,6 +993,15 @@ void sse_pmovmskb( struct x86_function *p,
    emit_modrm(p, dst, src);
 }
 
+void sse_movmskps( struct x86_function *p,
+                   struct x86_reg dst,
+                   struct x86_reg src)
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x50);
+   emit_modrm(p, dst, src);
+}
+
 /***********************************************************************
  * SSE2 instructions
  */
@@ -1558,6 +1680,21 @@ void mmx_movq( struct x86_function *p,
  */
 
 
+void x86_cdecl_caller_push_regs( struct x86_function *p )
+{
+   x86_push(p, x86_make_reg(file_REG32, reg_AX));
+   x86_push(p, x86_make_reg(file_REG32, reg_CX));
+   x86_push(p, x86_make_reg(file_REG32, reg_DX));
+}
+
+void x86_cdecl_caller_pop_regs( struct x86_function *p )
+{
+   x86_pop(p, x86_make_reg(file_REG32, reg_DX));
+   x86_pop(p, x86_make_reg(file_REG32, reg_CX));
+   x86_pop(p, x86_make_reg(file_REG32, reg_AX));
+}
+
+
 /* Retreive a reference to one of the function arguments, taking into
  * account any push/pop activity:
  */