From 1ba10e5ccf5cd0c990922e982e1e9bc6be48a5e4 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Wed, 21 May 2008 09:44:16 +0100
Subject: draw: add aos vertex shader varient

---
 src/gallium/auxiliary/draw/draw_vs_aos.h | 181 +++++++++++++++++++++++++++++++
 1 file changed, 181 insertions(+)
 create mode 100644 src/gallium/auxiliary/draw/draw_vs_aos.h

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
new file mode 100644
index 0000000000..1d8a055a90
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -0,0 +1,181 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef DRAW_VS_AOS_H
+#define DRAW_VS_AOS_H
+
+
+struct tgsi_token;
+struct x86_function;
+
+#include "pipe/p_state.h"
+#include "rtasm/rtasm_x86sse.h"
+
+
+
+
+
+#define X    0
+#define Y    1
+#define Z    2
+#define W    3
+
+#define MAX_INPUTS     PIPE_MAX_ATTRIBS
+#define MAX_OUTPUTS    PIPE_MAX_ATTRIBS
+#define MAX_TEMPS      PIPE_MAX_ATTRIBS /* say */
+#define MAX_CONSTANTS  PIPE_MAX_ATTRIBS /* say */
+#define MAX_IMMEDIATES PIPE_MAX_ATTRIBS /* say */
+#define MAX_INTERNALS  4
+
+#define AOS_FILE_INTERNAL TGSI_FILE_COUNT
+
+/* This is the temporary storage used by all the aos_sse vs varients.
+ * Create one per context and reuse by passing a pointer in at
+ * vs_varient creation??
+ */
+struct aos_machine {
+   float input    [MAX_INPUTS    ][4];
+   float output   [MAX_OUTPUTS   ][4];
+   float temp     [MAX_TEMPS     ][4];
+   float constant [MAX_CONSTANTS ][4]; /* fixme -- should just be a pointer */
+   float immediate[MAX_IMMEDIATES][4]; /* fixme -- should just be a pointer */
+   float internal [MAX_INTERNALS ][4];
+
+   unsigned fpu_round_nearest;
+   unsigned fpu_round_neg_inf;
+
+   struct {
+      const void *input_ptr;
+      unsigned input_stride;
+
+      unsigned output_offset;
+   } attrib[PIPE_MAX_ATTRIBS];
+};
+
+
+
+
+struct aos_compilation {
+   struct x86_function *func;
+   struct draw_vs_varient_aos_sse *vaos;
+
+   unsigned insn_counter;
+   unsigned num_immediates;
+
+   struct {
+      unsigned idx:16;
+      unsigned file:8;
+      unsigned dirty:8;
+      unsigned last_used;
+   } xmm[8];
+
+
+   boolean input_fetched[PIPE_MAX_ATTRIBS];
+   unsigned output_last_write[PIPE_MAX_ATTRIBS];
+
+   boolean have_sse2;
+   boolean error;
+   short fpucntl;
+
+   /* these are actually known values, but putting them in a struct
+    * like this is helpful to keep them in sync across the file.
+    */
+   struct x86_reg tmp_EAX;
+   struct x86_reg idx_EBX;     /* either start+i or &elt[i] */
+   struct x86_reg outbuf_ECX;
+   struct x86_reg machine_EDX;
+   struct x86_reg count_ESI;    /* decrements to zero */
+};
+
+struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp );
+void aos_release_xmm_reg( struct aos_compilation *cp, unsigned idx );
+
+void aos_adopt_xmm_reg( struct aos_compilation *cp,
+                        struct x86_reg reg,
+                        unsigned file,
+                        unsigned idx,
+                        unsigned dirty );
+
+struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, 
+                                   unsigned file,
+                                   unsigned idx );
+
+boolean aos_fetch_inputs( struct aos_compilation *cp,
+                          boolean linear );
+
+boolean aos_emit_outputs( struct aos_compilation *cp );
+
+
+#define IMM_ONES     0              /* 1, 1,1,1 */
+#define IMM_NEGS     1              /* 1,-1,0,0 */
+#define IMM_IDENTITY 2              /* 0, 0,0,1 */
+#define IMM_INV_255  3              /* 1/255, 1/255, 1/255, 1/255 */
+#define IMM_255      4              /* 255, 255, 255, 255 */
+
+struct x86_reg aos_get_internal( struct aos_compilation *cp,
+                                 unsigned imm );
+
+
+#define ERROR(cp, msg)                                                  \
+do {                                                                    \
+   debug_printf("%s: x86 translation failed: %s\n", __FUNCTION__, msg); \
+   cp->error = 1;                                                       \
+   assert(0);                                                           \
+} while (0)
+
+
+
+
+
+
+struct draw_vs_varient_aos_sse {
+   struct draw_vs_varient base;
+   struct draw_context *draw;
+
+#if 0
+   struct {
+      const void *ptr;
+      unsigned stride;
+   } attrib[PIPE_MAX_ATTRIBS];
+#endif
+
+   struct aos_machine *machine; /* XXX: temporarily unshared */
+
+   vsv_run_linear_func gen_run_linear;
+   vsv_run_elts_func gen_run_elts;
+
+
+   struct x86_function func[2];
+};
+
+
+
+#endif 
+
-- 
cgit v1.2.3


From 889473b3f5a216bd753c357974d6bae29fe3c41d Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Wed, 21 May 2008 20:28:56 +0100
Subject: draw: add viewport to varient state

---
 .../auxiliary/draw/draw_pt_fetch_shade_emit.c      |  9 +++-
 src/gallium/auxiliary/draw/draw_vs.h               |  8 +++-
 src/gallium/auxiliary/draw/draw_vs_aos.c           | 50 ++++++++++++++++++++++
 src/gallium/auxiliary/draw/draw_vs_aos.h           |  9 +++-
 src/gallium/auxiliary/draw/draw_vs_sse.c           |  4 +-
 src/gallium/auxiliary/draw/draw_vs_varient.c       | 10 +++++
 6 files changed, 84 insertions(+), 6 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 74945dcfe9..984fbb6767 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -95,10 +95,14 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
    
 
 
+   fse->key.output_stride = vinfo->size * 4;
    fse->key.nr_elements = MAX2(num_vs_outputs,     /* outputs - translate to hw format */
                                num_vs_inputs);     /* inputs - fetch from api format */
 
-   fse->key.output_stride = vinfo->size * 4;
+   fse->key.viewport = 1;
+   fse->key.clip = 0;
+   fse->key.pad = 0;
+
    memset(fse->key.element, 0, 
           fse->key.nr_elements * sizeof(fse->key.element[0]));
 
@@ -211,6 +215,9 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
    fse->active->set_constants( fse->active,
                                (const float (*)[4])draw->pt.user.constants );
 
+   fse->active->set_viewport( fse->active,
+                              &draw->viewport );
+
    //return TRUE;
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index 5a8d0da06d..ff3e19b2a8 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -58,7 +58,10 @@ struct draw_vs_element {
 
 struct draw_vs_varient_key {
    unsigned output_stride;
-   unsigned nr_elements;
+   unsigned nr_elements:16;
+   unsigned viewport:1;
+   unsigned clip:1;
+   unsigned pad:14;
    struct draw_vs_element element[PIPE_MAX_ATTRIBS];
 };
 
@@ -88,6 +91,9 @@ struct draw_vs_varient {
    void (*set_constants)( struct draw_vs_varient *,
                           const float (*constants)[4] );
 
+   void (*set_viewport)( struct draw_vs_varient *,
+                         const struct pipe_viewport_state * );
+
    void (PIPE_CDECL *run_linear)( struct draw_vs_varient *shader,
                                   unsigned start,
                                   unsigned count,
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 620f5e3592..b8e66e8b78 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -1401,6 +1401,37 @@ emit_instruction( struct aos_compilation *cp,
    }
 }
 
+
+static boolean emit_viewport( struct aos_compilation *cp )
+{
+   struct x86_reg pos = aos_get_shader_reg(cp, 
+                                           TGSI_FILE_OUTPUT, 
+                                           0);
+
+   struct x86_reg scale = x86_make_disp(cp->machine_EDX, 
+                                        Offset(struct aos_machine, scale));
+
+   struct x86_reg translate = x86_make_disp(cp->machine_EDX, 
+                                        Offset(struct aos_machine, translate));
+
+   if (pos.file != file_XMM) {
+      struct x86_reg dst = aos_get_xmm_reg(cp);
+      sse_movups(cp->func, dst, pos);
+      pos = dst;
+   }
+
+   sse_mulps(cp->func, pos, scale);
+   sse_addps(cp->func, pos, translate);
+
+   aos_adopt_xmm_reg( cp,
+                      pos,
+                      TGSI_FILE_OUTPUT,
+                      0,
+                      TRUE );
+   return TRUE;
+}
+
+
 static boolean note_immediate( struct aos_compilation *cp,
                                struct tgsi_full_immediate *imm )
 {
@@ -1540,6 +1571,10 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
       if (cp.error)
          goto fail;
 
+      if (cp.vaos->base.key.viewport) {
+         emit_viewport(&cp);
+      }
+
       /* Emit output...  TODO: do this eagerly after the last write to a
        * given output.
        */
@@ -1665,11 +1700,25 @@ static void vaos_set_constants( struct draw_vs_varient *varient,
 }
 
 
+static void vaos_set_viewport( struct draw_vs_varient *varient,
+                               const struct pipe_viewport_state *viewport )
+{
+   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+
+   memcpy(vaos->machine->scale, viewport->scale, 4 * sizeof(float));
+   memcpy(vaos->machine->translate, viewport->translate, 4 * sizeof(float));
+}
+
+
+
 static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
                                                  const struct draw_vs_varient_key *key )
 {
    struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse);
 
+   if (key->clip)
+      return NULL;
+
    if (!vaos)
       goto fail;
    
@@ -1677,6 +1726,7 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    vaos->base.vs = vs;
    vaos->base.set_input = vaos_set_buffer;
    vaos->base.set_constants = vaos_set_constants;
+   vaos->base.set_viewport = vaos_set_viewport;
    vaos->base.destroy = vaos_destroy;
    vaos->base.run_linear = vaos_run_linear;
    vaos->base.run_elts = vaos_run_elts;
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 1d8a055a90..16fef6451c 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -68,8 +68,13 @@ struct aos_machine {
    float immediate[MAX_IMMEDIATES][4]; /* fixme -- should just be a pointer */
    float internal [MAX_INTERNALS ][4];
 
-   unsigned fpu_round_nearest;
-   unsigned fpu_round_neg_inf;
+   float scale[4];              /* viewport */
+   float translate[4];          /* viewport */
+
+   ushort fpu_round_nearest;
+   ushort fpu_round_neg_inf;
+   ushort fpu_restore;
+   ushort fpucntl;              /* one of FPU_* above */
 
    struct {
       const void *input_ptr;
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 0581c3042f..7781782ae8 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -156,8 +156,8 @@ draw_create_vs_sse(struct draw_context *draw,
    tgsi_scan_shader(templ->tokens, &vs->base.info);
 
    vs->base.draw = draw;
-   vs->base.create_varient = draw_vs_varient_generic;
-//   vs->base.create_varient = draw_vs_varient_aos_sse;
+   vs->base.create_varient = draw_vs_varient_aos_sse;
+//   vs->base.create_varient = draw_vs_varient_generic;
    vs->base.prepare = vs_sse_prepare;
    vs->base.run_linear = vs_sse_run_linear;
    vs->base.delete = vs_sse_delete;
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index d27b0f6187..f6f621a748 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -167,6 +167,12 @@ static void vsvg_run_linear( struct draw_vs_varient *varient,
 
 
 
+
+static void vsvg_set_viewport( struct draw_vs_varient *varient,
+                               const struct pipe_viewport_state *viewport )
+{
+}
+
 static void vsvg_destroy( struct draw_vs_varient *varient )
 {
    FREE(varient);
@@ -179,6 +185,9 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    unsigned i;
    struct translate_key fetch, emit;
 
+   if (key->viewport || key->clip)
+      return NULL;
+
    struct draw_vs_varient_generic *vsvg = CALLOC_STRUCT( draw_vs_varient_generic );
    if (vsvg == NULL)
       return NULL;
@@ -187,6 +196,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    vsvg->base.vs = vs;
    vsvg->base.set_input     = vsvg_set_input;
    vsvg->base.set_constants = vsvg_set_constants;
+   vsvg->base.set_viewport  = vsvg_set_viewport;
    vsvg->base.run_elts      = vsvg_run_elts;
    vsvg->base.run_linear    = vsvg_run_linear;
    vsvg->base.destroy       = vsvg_destroy;
-- 
cgit v1.2.3


From 194a7be28f6eed502f2475d9a637cb3610ca75f6 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Wed, 21 May 2008 20:31:08 +0100
Subject: draw: fix vs aos internal/machine state

---
 src/gallium/auxiliary/draw/draw_vs_aos.c | 59 ++++++++++++++++++++++++++++++--
 src/gallium/auxiliary/draw/draw_vs_aos.h |  9 +++--
 2 files changed, 63 insertions(+), 5 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index b8e66e8b78..67761f881d 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -83,7 +83,7 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
       return x86_make_disp(ptr, Offset(struct aos_machine, constant[idx]));
 
    case AOS_FILE_INTERNAL:
-      return x86_make_disp(ptr, Offset(struct aos_machine, immediate[idx]));
+      return x86_make_disp(ptr, Offset(struct aos_machine, internal[idx]));
 
    default:
       ERROR(cp, "unknown reg file");
@@ -97,9 +97,63 @@ struct x86_reg aos_get_internal( struct aos_compilation *cp,
 {
    return get_reg_ptr( cp,
                        AOS_FILE_INTERNAL, 
-                       imm + 1 );
+                       imm );
+}
+
+#define X87_CW_EXCEPTION_INV_OP       (1<<0)
+#define X87_CW_EXCEPTION_DENORM_OP    (1<<1)
+#define X87_CW_EXCEPTION_ZERO_DIVIDE  (1<<2)
+#define X87_CW_EXCEPTION_OVERFLOW     (1<<3)
+#define X87_CW_EXCEPTION_UNDERFLOW    (1<<4)
+#define X87_CW_EXCEPTION_PRECISION    (1<<5)
+#define X87_CW_PRECISION_SINGLE       (0<<8)
+#define X87_CW_PRECISION_RESERVED     (1<<8)
+#define X87_CW_PRECISION_DOUBLE       (2<<8)
+#define X87_CW_PRECISION_DOUBLE_EXT   (3<<8)
+#define X87_CW_PRECISION_MASK         (3<<8)
+#define X87_CW_ROUND_NEAREST          (0<<10)
+#define X87_CW_ROUND_DOWN             (1<<10)
+#define X87_CW_ROUND_UP               (2<<10)
+#define X87_CW_ROUND_ZERO             (3<<10)
+#define X87_CW_ROUND_MASK             (3<<10)
+#define X87_CW_INFINITY               (1<<12)
+
+static void init_internals( struct aos_machine *machine )
+{
+   float inv = 1.0f/255.0f;
+   float f255 = 255.0f;
+
+   ASSIGN_4V(machine->internal[IMM_ONES],      1.0f,  1.0f,  1.0f,  1.0f);
+   ASSIGN_4V(machine->internal[IMM_NEGS],     -1.0f, -1.0f, -1.0f, -1.0f);
+   ASSIGN_4V(machine->internal[IMM_IDENTITY],  0.0f,  0.0f,  0.0f,  1.0f);
+   ASSIGN_4V(machine->internal[IMM_INV_255],   inv,   inv,   inv,   inv);
+   ASSIGN_4V(machine->internal[IMM_255],       f255,   f255,   f255,   f255);
+
+
+   machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP |
+                               X87_CW_EXCEPTION_DENORM_OP |
+                               X87_CW_EXCEPTION_ZERO_DIVIDE |
+                               X87_CW_EXCEPTION_OVERFLOW |
+                               X87_CW_EXCEPTION_UNDERFLOW |
+                               X87_CW_EXCEPTION_PRECISION |
+                               (1<<6) |
+                               X87_CW_ROUND_NEAREST |
+                               X87_CW_PRECISION_DOUBLE_EXT);
+
+   assert(machine->fpu_rnd_nearest == 0x37f);
+                               
+   machine->fpu_rnd_neg_inf = (X87_CW_EXCEPTION_INV_OP |
+                               X87_CW_EXCEPTION_DENORM_OP |
+                               X87_CW_EXCEPTION_ZERO_DIVIDE |
+                               X87_CW_EXCEPTION_OVERFLOW |
+                               X87_CW_EXCEPTION_UNDERFLOW |
+                               X87_CW_EXCEPTION_PRECISION |
+                               (1<<6) |
+                               X87_CW_ROUND_DOWN |
+                               X87_CW_PRECISION_DOUBLE_EXT);
 }
 
+
 static void spill( struct aos_compilation *cp, unsigned idx )
 {
    if (!cp->xmm[idx].dirty ||
@@ -1736,6 +1790,7 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
       goto fail;
    
    memset(vaos->machine, 0, sizeof(struct aos_machine));
+   init_internals(vaos->machine);
 
    tgsi_dump(vs->state.tokens, 0);
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 16fef6451c..c2afd4e9a0 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -52,10 +52,13 @@ struct x86_function;
 #define MAX_TEMPS      PIPE_MAX_ATTRIBS /* say */
 #define MAX_CONSTANTS  PIPE_MAX_ATTRIBS /* say */
 #define MAX_IMMEDIATES PIPE_MAX_ATTRIBS /* say */
-#define MAX_INTERNALS  4
+#define MAX_INTERNALS  8
 
 #define AOS_FILE_INTERNAL TGSI_FILE_COUNT
 
+#define FPU_RND_NEG    1
+#define FPU_RND_NEAREST 2
+
 /* This is the temporary storage used by all the aos_sse vs varients.
  * Create one per context and reuse by passing a pointer in at
  * vs_varient creation??
@@ -71,8 +74,8 @@ struct aos_machine {
    float scale[4];              /* viewport */
    float translate[4];          /* viewport */
 
-   ushort fpu_round_nearest;
-   ushort fpu_round_neg_inf;
+   ushort fpu_rnd_nearest;
+   ushort fpu_rnd_neg_inf;
    ushort fpu_restore;
    ushort fpucntl;              /* one of FPU_* above */
 
-- 
cgit v1.2.3


From c684ffa02d8d43ee04b99ee63ccd1adb66e81c1a Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Thu, 22 May 2008 13:41:49 +0100
Subject: draw: clean up internal immediates in aos sse

---
 src/gallium/auxiliary/draw/draw_vs_aos.c    | 64 +++++++++++++++++++++--------
 src/gallium/auxiliary/draw/draw_vs_aos.h    |  5 ++-
 src/gallium/auxiliary/draw/draw_vs_aos_io.c | 10 ++---
 3 files changed, 55 insertions(+), 24 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index fde92c7226..0b8600696a 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -44,12 +44,6 @@
 #ifdef PIPE_ARCH_X86
 
 
-#define DISASSEM 0
-
-
-
-
-
 static INLINE boolean eq( struct x86_reg a,
 			    struct x86_reg b )
 {
@@ -92,13 +86,6 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
 }
 		
 
-struct x86_reg aos_get_internal( struct aos_compilation *cp,
-                                 unsigned imm )
-{
-   return get_reg_ptr( cp,
-                       AOS_FILE_INTERNAL, 
-                       imm );
-}
 
 #define X87_CW_EXCEPTION_INV_OP       (1<<0)
 #define X87_CW_EXCEPTION_DENORM_OP    (1<<1)
@@ -123,6 +110,9 @@ static void init_internals( struct aos_machine *machine )
    float inv = 1.0f/255.0f;
    float f255 = 255.0f;
 
+   ASSIGN_4V(machine->internal[IMM_SWZ],       1.0f,  -1.0f,  0.0f, 1.0f);
+   *(unsigned *)&machine->internal[IMM_SWZ][3] = 0xffffffff;
+
    ASSIGN_4V(machine->internal[IMM_ONES],      1.0f,  1.0f,  1.0f,  1.0f);
    ASSIGN_4V(machine->internal[IMM_NEGS],     -1.0f, -1.0f, -1.0f, -1.0f);
    ASSIGN_4V(machine->internal[IMM_IDENTITY],  0.0f,  0.0f,  0.0f,  1.0f);
@@ -337,6 +327,39 @@ struct x86_reg aos_get_shader_reg( struct aos_compilation *cp,
 
 
 
+static struct x86_reg aos_get_shader_reg_xmm( struct aos_compilation *cp, 
+                                              unsigned file,
+                                              unsigned idx )
+{
+   struct x86_reg reg = aos_get_shader_reg( cp, file, idx );
+
+   if (reg.file != file_XMM) {
+      struct x86_reg tmp = aos_get_xmm_reg(cp);
+      sse_movups(cp->func, tmp, reg);
+      aos_adopt_xmm_reg( cp, tmp, file, idx, FALSE );
+      reg = tmp;
+   }
+
+   return reg;
+}
+
+
+
+struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp,
+                                     unsigned imm )
+{
+   return aos_get_shader_reg_xmm( cp, AOS_FILE_INTERNAL, imm );
+}
+
+
+struct x86_reg aos_get_internal( struct aos_compilation *cp,
+                                 unsigned imm )
+{
+   return aos_get_shader_reg( cp, AOS_FILE_INTERNAL, imm );
+}
+
+
+
 
 
 /* Emulate pshufd insn in regular SSE, if necessary:
@@ -461,15 +484,15 @@ static struct x86_reg fetch_src( struct aos_compilation *cp,
          arg0 = dst;
       }
 
-      if (negs) {
-         struct x86_reg imm_negs = aos_get_internal(cp, IMM_NEGS);
+      if (negs && negs != 0xf) {
+         struct x86_reg imm_swz = aos_get_internal_xmm(cp, IMM_SWZ);
          struct x86_reg tmp = aos_get_xmm_reg(cp);
 
          /* Load 1,-1,0,0
           * Use neg as arg to pshufd
           * Multiply
           */
-         emit_pshufd(cp, tmp, imm_negs, 
+         emit_pshufd(cp, tmp, imm_swz, 
                      SHUF((negs & 1) ? 1 : 0,
                           (negs & 2) ? 1 : 0,
                           (negs & 4) ? 1 : 0,
@@ -479,12 +502,17 @@ static struct x86_reg fetch_src( struct aos_compilation *cp,
          aos_release_xmm_reg(cp, tmp.idx);
          arg0 = dst;
       }
+      else if (negs) {
+         struct x86_reg imm_negs = aos_get_internal_xmm(cp, IMM_NEGS);
+         sse_mulps(cp->func, dst, imm_negs);
+         arg0 = dst;
+      }
+
 
       if (abs && abs != 0xf) {
          ERROR(cp, "unsupported partial abs");
       }
-
-      if (abs) {
+      else if (abs) {
          struct x86_reg neg = aos_get_internal(cp, IMM_NEGS);
          struct x86_reg tmp = aos_get_xmm_reg(cp);
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index c2afd4e9a0..efdc9a38f4 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -142,13 +142,16 @@ boolean aos_emit_outputs( struct aos_compilation *cp );
 
 
 #define IMM_ONES     0              /* 1, 1,1,1 */
-#define IMM_NEGS     1              /* 1,-1,0,0 */
+#define IMM_SWZ      1              /* 1,-1,0, 0xffffffff */
 #define IMM_IDENTITY 2              /* 0, 0,0,1 */
 #define IMM_INV_255  3              /* 1/255, 1/255, 1/255, 1/255 */
 #define IMM_255      4              /* 255, 255, 255, 255 */
+#define IMM_NEGS     5              /* -1,-1,-1,-1 */
 
 struct x86_reg aos_get_internal( struct aos_compilation *cp,
                                  unsigned imm );
+struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp,
+                                     unsigned imm );
 
 
 #define ERROR(cp, msg)                                                  \
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index 72b2b3d11d..0dda9df97d 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -54,7 +54,7 @@ static void emit_load_R32G32B32( struct aos_compilation *cp,
 				 struct x86_reg src_ptr )
 {
    sse_movss(cp->func, data, x86_make_disp(src_ptr, 8));
-   sse_shufps(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) );
+   sse_shufps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) );
    sse_shufps(cp->func, data, data, SHUF(Y,Z,X,W) );
    sse_movlps(cp->func, data, src_ptr);
 }
@@ -63,7 +63,7 @@ static void emit_load_R32G32( struct aos_compilation *cp,
 			   struct x86_reg data,
 			   struct x86_reg src_ptr )
 {
-   sse_movups(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ) );
+   sse_movups(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) );
    sse_movlps(cp->func, data, src_ptr);
 }
 
@@ -73,7 +73,7 @@ static void emit_load_R32( struct aos_compilation *cp,
 			   struct x86_reg src_ptr )
 {
    sse_movss(cp->func, data, src_ptr);
-   sse_orps(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ) );
+   sse_orps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) );
 }
 
 
@@ -82,8 +82,8 @@ static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp,
 				       struct x86_reg src_ptr )
 {
    sse_movss(cp->func, data, src_ptr);
-   sse2_punpcklbw(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ));
-   sse2_punpcklbw(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ));
+   sse2_punpcklbw(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ));
+   sse2_punpcklbw(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ));
    sse2_cvtdq2ps(cp->func, data, data);
    sse_mulps(cp->func, data, aos_get_internal(cp, IMM_INV_255));
 }
-- 
cgit v1.2.3


From 6172f1295cf812108d8ceba15a83ba87880360d3 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Sat, 24 May 2008 13:22:29 +0100
Subject: draw: add a debug-print which can be called from inside generated
 shaders

---
 src/gallium/auxiliary/draw/draw_vs_aos.c | 67 ++++++++++++++++++++++++++++++++
 src/gallium/auxiliary/draw/draw_vs_aos.h |  1 +
 2 files changed, 68 insertions(+)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index b0c3ac49d2..aa119f242e 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -752,7 +752,63 @@ static void x87_emit_ex2( struct aos_compilation *cp )
       
 }
 
+static void PIPE_CDECL print_reg( const char *msg,
+                                  const float *reg )
+{
+   debug_printf("%s: %f %f %f %f\n", msg, reg[0], reg[1], reg[2], reg[3]);
+}
+
+static void emit_print( struct aos_compilation *cp,
+                        const char *message, /* must point to a static string! */
+                        unsigned file,
+                        unsigned idx )
+{
+   struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
+   struct x86_reg arg = get_reg_ptr( cp, file, idx );
+   unsigned i;
+
+   /* There shouldn't be anything on the x87 stack.  Can add this
+    * capacity later if need be.
+    */
+   assert(cp->func->x87_stack == 0);
+
+   /* For absolute correctness, need to spill/invalidate all XMM regs
+    * too.  We're obviously not concerned about performance on this
+    * debug path, so here goes:
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
 
+      aos_release_xmm_reg(cp, i);
+   }
+
+   /* Push caller-save (ie scratch) regs.  
+    */
+   x86_cdecl_caller_push_regs( cp->func );
+
+
+   /* Push the arguments:
+    */
+   x86_lea( cp->func, ecx, arg );
+   x86_push( cp->func, ecx );
+   x86_push_imm32( cp->func, (int)message );
+
+   /* Call the helper.  Could call debug_printf directly, but
+    * print_reg is a nice place to put a breakpoint if need be.
+    */
+   x86_mov_reg_imm( cp->func, ecx, (int)print_reg );
+   x86_call( cp->func, ecx );
+   x86_pop( cp->func, ecx );
+   x86_pop( cp->func, ecx );
+
+   /* Pop caller-save regs 
+    */
+   x86_cdecl_caller_pop_regs( cp->func );
+
+   /* Done... 
+    */
+}
 
 /**
  * The traditional instructions.  All operate on internal registers
@@ -1798,6 +1854,17 @@ static void vaos_set_constants( struct draw_vs_varient *varient,
    memcpy(vaos->machine->constant,
           constants,
           (vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1) * 4 * sizeof(float));
+
+#if 0
+   unsigned i;
+   for (i =0; i < vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1; i++)
+      debug_printf("state %d: %f %f %f %f\n",
+                   i, 
+                   constants[i][0],
+                   constants[i][1],
+                   constants[i][2],
+                   constants[i][3]);
+#endif
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index efdc9a38f4..a0680ec63d 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -96,6 +96,7 @@ struct aos_compilation {
 
    unsigned insn_counter;
    unsigned num_immediates;
+   unsigned count;
 
    struct {
       unsigned idx:16;
-- 
cgit v1.2.3


From 86e529ad90411d21bca3d70984b2db202e7a0cd6 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Sat, 24 May 2008 16:31:11 +0100
Subject: draw: use lookup tables to avoid calling pow() in LIT opcode

---
 src/gallium/auxiliary/draw/draw_vs_aos.c | 251 ++++++++++++++++++++++++++++++-
 src/gallium/auxiliary/draw/draw_vs_aos.h |  27 ++++
 2 files changed, 272 insertions(+), 6 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index aa119f242e..1fbb7088ca 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -105,8 +105,31 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
 #define X87_CW_ROUND_MASK             (3<<10)
 #define X87_CW_INFINITY               (1<<12)
 
+static void do_populate_lut( struct shine_tab *tab,
+                             float unclamped_exponent )
+{
+   const float epsilon = 1.0F / 256.0F;    
+   float exponent = CLAMP(unclamped_exponent, -(128.0F - epsilon), (128.0F - epsilon));
+   unsigned i;
+
+   tab->exponent = unclamped_exponent; /* for later comparison */
+   
+   tab->values[0] = 0;
+   if (exponent == 0) {
+      for (i = 1; i < 258; i++) {
+         tab->values[i] = 1.0;
+      }      
+   }
+   else {
+      for (i = 1; i < 258; i++) {
+         tab->values[i] = powf((float)i * epsilon, exponent);
+      }
+   }
+}
+
 static void init_internals( struct aos_machine *machine )
 {
+   unsigned i;
    float inv = 1.0f/255.0f;
    float f255 = 255.0f;
 
@@ -141,6 +164,9 @@ static void init_internals( struct aos_machine *machine )
                                (1<<6) |
                                X87_CW_ROUND_DOWN |
                                X87_CW_PRECISION_DOUBLE_EXT);
+
+   for (i = 0; i < MAX_SHINE_TAB; i++)
+      do_populate_lut( &machine->shine_tab[i], 1.0f );
 }
 
 
@@ -1132,26 +1158,231 @@ static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_inst
    return TRUE;
 }
 
+static PIPE_CDECL void do_lit( struct aos_machine *machine,
+                               float *result,
+                               const float *in,
+                               unsigned count )
+{
+   if (in[0] > 0) 
+   {
+      if (in[1] <= 0.0) 
+      {
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = 1.0;
+         result[3] = 1.0F;
+      }
+      else
+      {
+         const float epsilon = 1.0F / 256.0F;    
+         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = powf(in[1], exponent);
+         result[3] = 1.0;
+      }
+   }
+   else 
+   {
+      result[0] = 1.0F;
+      result[1] = 0.0;
+      result[2] = 0.0;
+      result[3] = 1.0F;
+   }
+}
+
+
+static PIPE_CDECL void do_lit_lut( struct aos_machine *machine,
+                                   float *result,
+                                   const float *in,
+                                   unsigned count )
+{
+   if (in[0] > 0) 
+   {
+      if (in[1] <= 0.0) 
+      {
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = 1.0;
+         result[3] = 1.0F;
+         return;
+      }
+      
+      if (machine->lit_info[count].shine_tab->exponent != in[3]) {
+         machine->lit_info[count].func = do_lit;
+         goto no_luck;
+      }
+
+      if (in[1] <= 1.0)
+      {
+         const float *tab = machine->lit_info[count].shine_tab->values;
+         float f = in[1] * 256;
+         int k = (int)f;
+         float frac = f - (float)k;
+         
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = tab[k] + frac*(tab[k+1]-tab[k]);
+         result[3] = 1.0;
+         return;
+      }
+      
+   no_luck:
+      {
+         const float epsilon = 1.0F / 256.0F;    
+         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = powf(in[1], exponent);
+         result[3] = 1.0;
+      }
+   }
+   else 
+   {
+      result[0] = 1.0F;
+      result[1] = 0.0;
+      result[2] = 0.0;
+      result[3] = 1.0F;
+   }
+}
+
+
+
+static void PIPE_CDECL populate_lut( struct aos_machine *machine,
+                                     float *result,
+                                     const float *in,
+                                     unsigned count )
+{
+   unsigned i, tab;
+
+   /* Search for an existing table for this value.  Note that without
+    * static analysis we don't really know if in[3] will be constant,
+    * but it usually is...
+    */
+   for (tab = 0; tab < 4; tab++) {
+      if (machine->shine_tab[tab].exponent == in[3]) {
+         goto found;
+      }
+   }
+
+   for (tab = 0, i = 1; i < 4; i++) {
+      if (machine->shine_tab[i].last_used < machine->shine_tab[tab].last_used)
+         tab = i;
+   }
+
+   if (machine->shine_tab[tab].last_used == machine->now) {
+      /* No unused tables (this is not a ffvertex program...).  Just
+       * call pow each time:
+       */
+      machine->lit_info[count].func = do_lit;
+      machine->lit_info[count].func( machine, result, in, count );
+      return;
+   }
+   else {
+      do_populate_lut( &machine->shine_tab[tab], in[3] );
+   }
+
+ found:
+   machine->shine_tab[tab].last_used = machine->now;
+   machine->lit_info[count].shine_tab = &machine->shine_tab[tab];
+   machine->lit_info[count].func = do_lit_lut;
+   machine->lit_info[count].func( machine, result, in, count );
+}
+
+
+
 
 
 static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
 {
-   struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); 
+   struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
    unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask;
+   unsigned lit_count = cp->lit_count++;
+   struct x86_reg result, arg0;
+   unsigned i;
+
+#if 1
+   /* For absolute correctness, need to spill/invalidate all XMM regs
+    * too.  
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+#endif
+
+   if (writemask != TGSI_WRITEMASK_XYZW) 
+      result = x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, tmp[0]));
+   else 
+      result = get_dst_ptr(cp, &op->FullDstRegisters[0]);    
+
+   
+   arg0 = fetch_src( cp, &op->FullSrcRegisters[0] );
+   if (arg0.file == file_XMM) {
+      struct x86_reg tmp = x86_make_disp(cp->machine_EDX, 
+                                         Offset(struct aos_machine, tmp[1]));
+      sse_movaps( cp->func, tmp, arg0 );
+      arg0 = tmp;
+   }
+                  
+      
+
+   /* Push caller-save (ie scratch) regs.  
+    */
+   x86_cdecl_caller_push_regs( cp->func );
+
+   /* Push the arguments:
+    */
+   x86_push_imm32( cp->func, lit_count );
+
+   x86_lea( cp->func, ecx, arg0 );
+   x86_push( cp->func, ecx );
 
+   x86_lea( cp->func, ecx, result );
+   x86_push( cp->func, ecx );
 
+   x86_push( cp->func, cp->machine_EDX );
 
+   if (lit_count < MAX_LIT_INFO) {
+      x86_mov( cp->func, ecx, x86_make_disp( cp->machine_EDX, 
+                                             Offset(struct aos_machine, lit_info) + 
+                                             lit_count * sizeof(struct lit_info) + 
+                                             Offset(struct lit_info, func)));
+   }
+   else {
+      x86_mov_reg_imm( cp->func, ecx, (int)do_lit );
+   }
+
+   x86_call( cp->func, ecx );
+            
+   x86_pop( cp->func, ecx );    /* fixme... */
+   x86_pop( cp->func, ecx );
+   x86_pop( cp->func, ecx );
+   x86_pop( cp->func, ecx );
+
+   x86_cdecl_caller_pop_regs( cp->func );
+
+   if (writemask != TGSI_WRITEMASK_XYZW) {
+      store_dest( cp, 
+                  &op->FullDstRegisters[0],
+                  get_xmm_clone( cp, result ) );
+   }
+
+   return TRUE;
+}
+
+   
+static boolean emit_inline_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); 
+   unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask;
 
    if (writemask & TGSI_WRITEMASK_YZ) {
       struct x86_reg st1 = x86_make_reg(file_x87, 1);
       struct x86_reg st2 = x86_make_reg(file_x87, 2);
 
-   
-      
-
       /* a1' = a1 <= 0 ? 1 : a1;  
-       *
-       * Note: use 1.0 to avoid passing zero to 
        */
       x87_fldz(cp->func);                           /* 1 0  */
 #if 1
@@ -1865,6 +2096,14 @@ static void vaos_set_constants( struct draw_vs_varient *varient,
                    constants[i][2],
                    constants[i][3]);
 #endif
+
+   {
+      unsigned i;
+      for (i = 0; i < MAX_LIT_INFO; i++) {
+         vaos->machine->lit_info[i].func = populate_lut;
+         vaos->machine->now++;
+      }
+   }
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index a0680ec63d..c08c73d4bc 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -59,6 +59,25 @@ struct x86_function;
 #define FPU_RND_NEG    1
 #define FPU_RND_NEAREST 2
 
+struct aos_machine;
+typedef void PIPE_CDECL (*lit_func)( struct aos_machine *,
+                                    float *result,
+                                    const float *in,
+                                    unsigned count );
+struct shine_tab {
+   float exponent;
+   float values[258];
+   unsigned last_used;
+};
+
+struct lit_info {
+   lit_func func;
+   struct shine_tab *shine_tab;
+};
+
+#define MAX_SHINE_TAB    4
+#define MAX_LIT_INFO     16
+
 /* This is the temporary storage used by all the aos_sse vs varients.
  * Create one per context and reuse by passing a pointer in at
  * vs_varient creation??
@@ -74,6 +93,13 @@ struct aos_machine {
    float scale[4];              /* viewport */
    float translate[4];          /* viewport */
 
+   float tmp[2][4];             /* scratch space for LIT */
+
+   struct shine_tab shine_tab[MAX_SHINE_TAB];
+   struct lit_info  lit_info[MAX_LIT_INFO];
+   unsigned now;
+   
+
    ushort fpu_rnd_nearest;
    ushort fpu_rnd_neg_inf;
    ushort fpu_restore;
@@ -97,6 +123,7 @@ struct aos_compilation {
    unsigned insn_counter;
    unsigned num_immediates;
    unsigned count;
+   unsigned lit_count;
 
    struct {
       unsigned idx:16;
-- 
cgit v1.2.3


From 351eca365c0ba488000c3826d5093de6170381e4 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Mon, 26 May 2008 11:03:00 +0100
Subject: draw: extend precision in RSQ opcode

---
 src/gallium/auxiliary/draw/draw_vs_aos.c | 48 ++++++++++++++++++--------------
 src/gallium/auxiliary/draw/draw_vs_aos.h |  1 +
 2 files changed, 28 insertions(+), 21 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 916203c66b..1622358ae1 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -140,7 +140,8 @@ static void init_internals( struct aos_machine *machine )
    ASSIGN_4V(machine->internal[IMM_NEGS],     -1.0f, -1.0f, -1.0f, -1.0f);
    ASSIGN_4V(machine->internal[IMM_IDENTITY],  0.0f,  0.0f,  0.0f,  1.0f);
    ASSIGN_4V(machine->internal[IMM_INV_255],   inv,   inv,   inv,   inv);
-   ASSIGN_4V(machine->internal[IMM_255],       f255,   f255,   f255,   f255);
+   ASSIGN_4V(machine->internal[IMM_255],       f255,  f255,  f255,  f255);
+   ASSIGN_4V(machine->internal[IMM_RSQ],       -.5f,  1.5f,  0.0f,  0.0f);
 
 
    machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP |
@@ -1561,35 +1562,40 @@ static boolean emit_RCP( struct aos_compilation *cp, const struct tgsi_full_inst
  * 
  * x1 = 2 * rcpps(a) - a * rcpps(a) * rcpps(a)
  * x1 = 0.5 * rsqrtps(a) * [3.0 - (a * rsqrtps(a))* rsqrtps(a)]
+ * or:
+ *   x1 = rsqrtps(a) * [1.5 - .5 * a * rsqrtps(a) * rsqrtps(a)]
+ * 
  *
  * See: http://softwarecommunity.intel.com/articles/eng/1818.htm
  */
 static boolean emit_RSQ( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
 {
-   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
-   struct x86_reg dst = aos_get_xmm_reg(cp);
 
-   if (1) {
-      sse_rsqrtss(cp->func, dst, arg0);
+   if (0) {
+      struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+      struct x86_reg r = aos_get_xmm_reg(cp);
+      sse_rsqrtss(cp->func, r, arg0);
+      store_scalar_dest(cp, &op->FullDstRegisters[0], r);
+      return TRUE;
    }
    else {
-#if 0
-      /* Extend precision here...
-       */
-      sse_movaps(  func, dst,  get_temp( TGSI_EXEC_TEMP_HALF_I, TGSI_EXEC_TEMP_HALF_C ) );
-      sse_movaps(  func, tmp0, get_temp( TGSI_EXEC_TEMP_THREE_I, TGSI_EXEC_TEMP_THREE_C ) );
-
-      sse_rsqrtss( func, tmp1, src  ); /* rsqrtss(a) */
-      sse_mulss(   func, src,  tmp1 ); /* a * rsqrtss(a) */
-      sse_mulss(   func, dst,  tmp1 ); /* .5 * rsqrtss(a) */
-      sse_mulss(   func, src,  tmp1 ); /* a * rsqrtss(a) * rsqrtss(a) */
-      sse_subss(   func, tmp0, src  ); /* 3.0 - (a * rsqrtss(a) * rsqrtss(a)) */
-      sse_mulss(   func, dst,  tmp0 ); /* .5 * r * (3.0 - (a * r * r)) */
-#endif
-   }
+      struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+      struct x86_reg r = aos_get_xmm_reg(cp);
 
-   store_scalar_dest(cp, &op->FullDstRegisters[0], dst);
-   return TRUE;
+      struct x86_reg neg_half       = get_reg_ptr( cp, AOS_FILE_INTERNAL, IMM_RSQ );
+      struct x86_reg one_point_five = x86_make_disp( neg_half, 4 );
+      struct x86_reg src            = get_xmm_writable( cp, arg0 );
+      
+      sse_rsqrtss( cp->func, r, src  );             /* rsqrtss(a) */
+      sse_mulss(   cp->func, src, neg_half  );      /* -.5 * a */
+      sse_mulss(   cp->func, src,  r );             /* -.5 * a * r */
+      sse_mulss(   cp->func, src,  r );             /* -.5 * a * r * r */
+      sse_addss(   cp->func, src, one_point_five ); /* 1.5 - .5 * a * r * r */
+      sse_mulss(   cp->func, r,  src );             /* r * (1.5 - .5 * a * r * r) */
+
+      store_scalar_dest(cp, &op->FullDstRegisters[0], r);
+      return TRUE;
+   }
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index c08c73d4bc..fffe2e4658 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -175,6 +175,7 @@ boolean aos_emit_outputs( struct aos_compilation *cp );
 #define IMM_INV_255  3              /* 1/255, 1/255, 1/255, 1/255 */
 #define IMM_255      4              /* 255, 255, 255, 255 */
 #define IMM_NEGS     5              /* -1,-1,-1,-1 */
+#define IMM_RSQ      6              /* -.5,1.5,_,_ */
 
 struct x86_reg aos_get_internal( struct aos_compilation *cp,
                                  unsigned imm );
-- 
cgit v1.2.3


From 7b85ea19de09d4e7e077ca147528e90e52683690 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Tue, 27 May 2008 19:01:57 +0100
Subject: draw: support psize in vs_varient paths

Preserve the vinfo "EMIT_*" format descriptors in the varient key,
and deal with PSIZE directly in each implementation.
---
 .../auxiliary/draw/draw_pt_fetch_shade_emit.c      | 35 ++------------------
 src/gallium/auxiliary/draw/draw_vertex.h           | 21 ++++++++++++
 src/gallium/auxiliary/draw/draw_vs_aos.c           |  3 ++
 src/gallium/auxiliary/draw/draw_vs_aos.h           |  1 +
 src/gallium/auxiliary/draw/draw_vs_aos_io.c        | 37 ++++++++++++++--------
 src/gallium/auxiliary/draw/draw_vs_varient.c       | 24 +++++++++++---
 6 files changed, 71 insertions(+), 50 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 85d0bdfcab..729c7db999 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -77,7 +77,6 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
    unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
    const struct vertex_info *vinfo;
    unsigned i;
-   boolean need_psize = 0;
    
 
    if (!draw->render->set_primitive( draw->render, 
@@ -123,34 +122,24 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
 
       for (i = 0; i < vinfo->num_attribs; i++) {
          unsigned emit_sz = 0;
-         unsigned output_format = PIPE_FORMAT_NONE;
-         unsigned vs_output = vinfo->src_index[i];
 
          switch (vinfo->emit[i]) {
          case EMIT_4F:
-            output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
             emit_sz = 4 * sizeof(float);
             break;
          case EMIT_3F:
-            output_format = PIPE_FORMAT_R32G32B32_FLOAT;
             emit_sz = 3 * sizeof(float);
             break;
          case EMIT_2F:
-            output_format = PIPE_FORMAT_R32G32_FLOAT;
             emit_sz = 2 * sizeof(float);
             break;
          case EMIT_1F:
-            output_format = PIPE_FORMAT_R32_FLOAT;
             emit_sz = 1 * sizeof(float);
             break;
          case EMIT_1F_PSIZE:
-            need_psize = 1;
-            output_format = PIPE_FORMAT_R32_FLOAT;
             emit_sz = 1 * sizeof(float);
-            vs_output = vinfo->num_attribs + 1;
             break;
          case EMIT_4UB:
-            output_format = PIPE_FORMAT_B8G8R8A8_UNORM;
             emit_sz = 4 * sizeof(ubyte);
             break;
          default:
@@ -162,33 +151,15 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
           * numbers, not to positions in the hw vertex description --
           * that's handled by the output_offset field.
           */
-         fse->key.element[vs_output].out.format = output_format;
-         fse->key.element[vs_output].out.vs_output = vs_output;
-         fse->key.element[vs_output].out.offset = dst_offset;
+         fse->key.element[i].out.format = vinfo->emit[i];
+         fse->key.element[i].out.vs_output = vinfo->src_index[i];
+         fse->key.element[i].out.offset = dst_offset;
       
          dst_offset += emit_sz;
          assert(fse->key.output_stride >= dst_offset);
       }
    }
 
-   /* To make psize work, really need to tell the vertex shader to
-    * copy that value from input->output.  For 'translate' this was
-    * implicit for all elements.
-    */
-#if 0
-   if (need_psize) {
-      unsigned input = num_vs_inputs + 1;
-      const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
-      fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT;
-      fse->key.element[i].input_buffer = 0; //nr_buffers + 1;
-      fse->key.element[i].input_offset = 0; 
-
-      fse->key.nr_inputs += 1;
-      fse->key.nr_elements = MAX2(fse->key.nr_inputs,
-                                  fse->key.nr_outputs);
-      
-   }
-#endif
 
    /* Would normally look up a vertex shader and peruse its list of
     * varients somehow.  We omitted that step and put all the
diff --git a/src/gallium/auxiliary/draw/draw_vertex.h b/src/gallium/auxiliary/draw/draw_vertex.h
index 6d8bac5138..16c65c4317 100644
--- a/src/gallium/auxiliary/draw/draw_vertex.h
+++ b/src/gallium/auxiliary/draw/draw_vertex.h
@@ -109,4 +109,25 @@ extern void draw_compute_vertex_size(struct vertex_info *vinfo);
 void draw_dump_emitted_vertex(const struct vertex_info *vinfo, 
                               const uint8_t *data);
 
+
+static INLINE unsigned draw_translate_vinfo_format(unsigned format )
+{
+   switch (format) {
+   case EMIT_1F:
+   case EMIT_1F_PSIZE:
+      return PIPE_FORMAT_R32_FLOAT;
+   case EMIT_2F:
+      return PIPE_FORMAT_R32G32_FLOAT;
+   case EMIT_3F:
+      return PIPE_FORMAT_R32G32B32_FLOAT;
+   case EMIT_4F:
+      return PIPE_FORMAT_R32G32B32A32_FLOAT;
+   case EMIT_4UB:
+      return PIPE_FORMAT_R8G8B8A8_UNORM;
+   default:
+      return PIPE_FORMAT_NONE;
+   }
+}
+
+
 #endif /* DRAW_VERTEX_H */
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 1c63677e6e..d3770b2c53 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -2126,6 +2126,7 @@ static void vaos_run_elts( struct draw_vs_varient *varient,
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
+   vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
    vaos->gen_run_elts( varient,
                        elts,
                        count,
@@ -2139,6 +2140,7 @@ static void vaos_run_linear( struct draw_vs_varient *varient,
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
+   vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
    vaos->gen_run_linear( varient,
                          start,
                          count,
@@ -2204,6 +2206,7 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    vaos->base.run_linear = vaos_run_linear;
    vaos->base.run_elts = vaos_run_elts;
 
+   vaos->draw = vs->draw;
    vaos->machine = align_malloc( sizeof(struct aos_machine), 16 );
    if (!vaos->machine)
       goto fail;
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index fffe2e4658..b47413ff43 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -176,6 +176,7 @@ boolean aos_emit_outputs( struct aos_compilation *cp );
 #define IMM_255      4              /* 255, 255, 255, 255 */
 #define IMM_NEGS     5              /* -1,-1,-1,-1 */
 #define IMM_RSQ      6              /* -.5,1.5,_,_ */
+#define IMM_PSIZE    7              /* not really an immediate - updated each run */
 
 struct x86_reg aos_get_internal( struct aos_compilation *cp,
                                  unsigned imm );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index cebfaf6474..836110f382 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -33,6 +33,7 @@
 #include "tgsi/exec/tgsi_exec.h"
 #include "draw_vs.h"
 #include "draw_vs_aos.h"
+#include "draw_vertex.h"
 
 #include "rtasm/rtasm_x86sse.h"
 
@@ -249,24 +250,27 @@ static boolean emit_output( struct aos_compilation *cp,
                             unsigned format )
 {
    switch (format) {
-   case PIPE_FORMAT_R32_FLOAT:
+   case EMIT_1F:
+   case EMIT_1F_PSIZE:
       emit_store_R32(cp, ptr, dataXMM);
       break;
-   case PIPE_FORMAT_R32G32_FLOAT:
+   case EMIT_2F:
       emit_store_R32G32(cp, ptr, dataXMM);
       break;
-   case PIPE_FORMAT_R32G32B32_FLOAT:
+   case EMIT_3F:
       emit_store_R32G32B32(cp, ptr, dataXMM);
       break;
-   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+   case EMIT_4F:
       emit_store_R32G32B32A32(cp, ptr, dataXMM);
       break;
-   case PIPE_FORMAT_B8G8R8A8_UNORM:
-      emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
-      emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
-      break;
-   case PIPE_FORMAT_R8G8B8A8_UNORM:
-      emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
+   case EMIT_4UB:
+      if (1) {
+         emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
+         emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
+      }
+      else {
+         emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
+      }
       break;
    default:
       ERROR(cp, "unhandled output format");
@@ -287,9 +291,16 @@ boolean aos_emit_outputs( struct aos_compilation *cp )
       unsigned offset = cp->vaos->base.key.element[i].out.offset;
       unsigned vs_output = cp->vaos->base.key.element[i].out.vs_output;
 
-      struct x86_reg data = aos_get_shader_reg( cp, 
-                                                TGSI_FILE_OUTPUT,
-                                                vs_output );
+      struct x86_reg data;
+
+      if (format == EMIT_1F_PSIZE) {
+         data = aos_get_internal_xmm( cp, IMM_PSIZE );
+      }
+      else {
+         data = aos_get_shader_reg( cp, 
+                                    TGSI_FILE_OUTPUT,
+                                    vs_output );
+      }
 
       if (data.file != file_XMM) {
          struct x86_reg tmp = aos_get_xmm_reg( cp );
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index dab46e8eed..119a3a04b5 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -231,6 +231,10 @@ static void vsvg_run_linear( struct draw_vs_varient *varient,
                               output_buffer,
                               vsvg->base.key.output_stride );
 
+      vsvg->emit->set_buffer( vsvg->emit, 
+                              1,
+                              &vsvg->draw->rasterizer->point_size,
+                              0);
 
       vsvg->emit->run( vsvg->emit,
                        0, count,
@@ -293,11 +297,21 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    emit.nr_elements = key->nr_outputs;
    emit.output_stride = key->output_stride;
    for (i = 0; i < key->nr_outputs; i++) {
-      emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-      emit.element[i].input_buffer = 0;
-      emit.element[i].input_offset = i * 4 * sizeof(float);
-      emit.element[i].output_format = key->element[i].out.format;
-      emit.element[i].output_offset = key->element[i].out.offset;
+      if (key->element[i].out.format != EMIT_1F_PSIZE)
+      {      
+         emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+         emit.element[i].input_buffer = 0;
+         emit.element[i].input_offset = key->element[i].out.vs_output * 4 * sizeof(float);
+         emit.element[i].output_format = draw_translate_vinfo_format(key->element[i].out.format);
+         emit.element[i].output_offset = key->element[i].out.offset;
+      }
+      else {
+         emit.element[i].input_format = PIPE_FORMAT_R32_FLOAT;
+         emit.element[i].input_buffer = 1;
+         emit.element[i].input_offset = 0;
+         emit.element[i].output_format = PIPE_FORMAT_R32_FLOAT;
+         emit.element[i].output_offset = key->element[i].out.offset;
+      }
    }
 
    vsvg->fetch = draw_vs_get_fetch( vs->draw, &fetch );
-- 
cgit v1.2.3


From 364f75d9dbc2b8e19c884b9cc74676ab6cbee60d Mon Sep 17 00:00:00 2001
From: José Fonseca <jrfonseca@tungstengraphics.com>
Date: Wed, 28 May 2008 22:19:58 +0900
Subject: draw: Use PIPE_CDECL.

---
 src/gallium/auxiliary/draw/draw_vs_aos.c | 20 ++++++++++----------
 src/gallium/auxiliary/draw/draw_vs_aos.h |  2 +-
 2 files changed, 11 insertions(+), 11 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index d3770b2c53..0cd82ff599 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -1189,7 +1189,7 @@ static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_inst
    return TRUE;
 }
 
-static PIPE_CDECL void do_lit( struct aos_machine *machine,
+static void PIPE_CDECL do_lit( struct aos_machine *machine,
                                float *result,
                                const float *in,
                                unsigned count )
@@ -1223,7 +1223,7 @@ static PIPE_CDECL void do_lit( struct aos_machine *machine,
 }
 
 
-static PIPE_CDECL void do_lit_lut( struct aos_machine *machine,
+static void PIPE_CDECL do_lit_lut( struct aos_machine *machine,
                                    float *result,
                                    const float *in,
                                    unsigned count )
@@ -2119,10 +2119,10 @@ static void vaos_destroy( struct draw_vs_varient *varient )
    FREE(vaos);
 }
 
-static void vaos_run_elts( struct draw_vs_varient *varient,
-                           const unsigned *elts,
-                           unsigned count,
-                           void *output_buffer )
+static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
+                                      const unsigned *elts,
+                                      unsigned count,
+                                      void *output_buffer )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
@@ -2133,10 +2133,10 @@ static void vaos_run_elts( struct draw_vs_varient *varient,
                        output_buffer );
 }
 
-static void vaos_run_linear( struct draw_vs_varient *varient,
-                             unsigned start,
-                             unsigned count,
-                             void *output_buffer )
+static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
+                                        unsigned start,
+                                        unsigned count,
+                                        void *output_buffer )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index b47413ff43..837b32794f 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -60,7 +60,7 @@ struct x86_function;
 #define FPU_RND_NEAREST 2
 
 struct aos_machine;
-typedef void PIPE_CDECL (*lit_func)( struct aos_machine *,
+typedef void (PIPE_CDECL *lit_func)( struct aos_machine *,
                                     float *result,
                                     const float *in,
                                     unsigned count );
-- 
cgit v1.2.3


From 728d1f7f43b6db9f4f42c2d16ba223c492d1147d Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Wed, 28 May 2008 23:54:18 +0100
Subject: draw: enable FSE by default

---
 src/gallium/auxiliary/draw/draw_pt.c        |  11 ++-
 src/gallium/auxiliary/draw/draw_vs.h        |   4 ++
 src/gallium/auxiliary/draw/draw_vs_aos.c    | 105 +++++++++++++++++++++-------
 src/gallium/auxiliary/draw/draw_vs_aos.h    |  34 +++++----
 src/gallium/auxiliary/draw/draw_vs_aos_io.c |  25 ++++---
 src/gallium/auxiliary/draw/draw_vs_sse.c    |  14 ++--
 6 files changed, 129 insertions(+), 64 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 75f44d503e..d48c6c220d 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -75,7 +75,7 @@ draw_pt_arrays(struct draw_context *draw,
 
    if (opt == 0) 
       middle = draw->pt.middle.fetch_emit;
-   else if (opt == PT_SHADE && draw->pt.test_fse)
+   else if (opt == PT_SHADE)
       middle = draw->pt.middle.fetch_shade_emit;
    else
       middle = draw->pt.middle.general;
@@ -118,12 +118,9 @@ boolean draw_pt_init( struct draw_context *draw )
    if (!draw->pt.middle.fetch_emit)
       return FALSE;
 
-   if (draw->pt.test_fse) {
-      draw->pt.middle.fetch_shade_emit = draw_pt_middle_fse( draw );
-      if (!draw->pt.middle.fetch_shade_emit)
-         return FALSE;
-   }
-
+   draw->pt.middle.fetch_shade_emit = draw_pt_middle_fse( draw );
+   if (!draw->pt.middle.fetch_shade_emit)
+      return FALSE;
 
    draw->pt.middle.general = draw_pt_fetch_pipeline_or_emit( draw );
    if (!draw->pt.middle.general)
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index 01171bc23d..7aa0415baf 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -123,6 +123,10 @@ struct draw_vertex_shader {
 
    struct tgsi_shader_info info;
 
+   /* Extracted from shader:
+    */
+   const float (*immediates)[4];
+
    /* 
     */
    struct draw_vs_varient *varient[16];
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 0cd82ff599..9056785e7a 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -66,6 +66,37 @@ static INLINE boolean eq( struct x86_reg a,
 	   a.disp == b.disp);
 }
       
+struct x86_reg aos_get_x86( struct aos_compilation *cp,
+                            unsigned value )
+{
+   if (cp->ebp != value) {
+      unsigned offset;
+
+      switch (value) {
+      case X86_IMMEDIATES:
+         offset = Offset(struct aos_machine, immediates);
+         break;
+      case X86_CONSTANTS:
+         offset = Offset(struct aos_machine, constants);
+         break;
+      case X86_ATTRIBS:
+         offset = Offset(struct aos_machine, attrib);
+         break;
+      default:
+         assert(0);
+         offset = 0;
+      }
+
+      x86_mov(cp->func, cp->temp_EBP, 
+              x86_make_disp(cp->machine_EDX, offset));
+      /* x86_deref(x86_make_disp(cp->machine_EDX, offset))); */
+
+      cp->ebp = value;
+   }
+
+   return cp->temp_EBP;
+}
+
 
 static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
                                   unsigned file,
@@ -83,15 +114,15 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
    case TGSI_FILE_TEMPORARY:
       return x86_make_disp(ptr, Offset(struct aos_machine, temp[idx]));
 
-   case TGSI_FILE_IMMEDIATE:
-      return x86_make_disp(ptr, Offset(struct aos_machine, immediate[idx]));
-
-   case TGSI_FILE_CONSTANT:       
-      return x86_make_disp(ptr, Offset(struct aos_machine, constant[idx]));
-
    case AOS_FILE_INTERNAL:
       return x86_make_disp(ptr, Offset(struct aos_machine, internal[idx]));
 
+   case TGSI_FILE_IMMEDIATE: 
+      return x86_make_disp(aos_get_x86(cp, X86_IMMEDIATES), idx * 4 * sizeof(float));
+
+   case TGSI_FILE_CONSTANT: 
+      return x86_make_disp(aos_get_x86(cp, X86_CONSTANTS), idx * 4 * sizeof(float));
+
    default:
       ERROR(cp, "unknown reg file");
       return x86_make_reg(0,0);
@@ -1865,6 +1896,7 @@ static boolean emit_rhw_viewport( struct aos_compilation *cp )
 }
 
 
+#if 0
 static boolean note_immediate( struct aos_compilation *cp,
                                struct tgsi_full_immediate *imm )
 {
@@ -1877,6 +1909,7 @@ static boolean note_immediate( struct aos_compilation *cp,
 
    return TRUE;
 }
+#endif
 
 
 
@@ -1939,6 +1972,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    cp.outbuf_ECX    = x86_make_reg(file_REG32, reg_CX);
    cp.machine_EDX   = x86_make_reg(file_REG32, reg_DX);
    cp.count_ESI     = x86_make_reg(file_REG32, reg_SI);
+   cp.temp_EBP     = x86_make_reg(file_REG32, reg_BP);
 
    x86_init_func(cp.func);
 
@@ -1946,6 +1980,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
    x86_push(cp.func, cp.idx_EBX);
    x86_push(cp.func, cp.count_ESI);
+   x86_push(cp.func, cp.temp_EBP);
 
 
    /* Load arguments into regs:
@@ -1988,8 +2023,10 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
          switch (parse.FullToken.Token.Type) {
          case TGSI_TOKEN_TYPE_IMMEDIATE:
+#if 0
             if (!note_immediate( &cp, &parse.FullToken.FullImmediate ))
                goto fail;
+#endif
             break;
 
          case TGSI_TOKEN_TYPE_INSTRUCTION:
@@ -2072,6 +2109,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    if (cp.func->need_emms)
       mmx_emms(cp.func);
 
+   x86_pop(cp.func, cp.temp_EBP);
    x86_pop(cp.func, cp.count_ESI);
    x86_pop(cp.func, cp.idx_EBX);
 
@@ -2098,26 +2136,14 @@ static void vaos_set_buffer( struct draw_vs_varient *varient,
 
    for (i = 0; i < vaos->base.key.nr_inputs; i++) {
       if (vaos->base.key.element[i].in.buffer == buf) {
-         vaos->machine->attrib[i].input_ptr = ((char *)ptr +
-                                               vaos->base.key.element[i].in.offset);
-         vaos->machine->attrib[i].input_stride = stride;
+         vaos->attrib[i].input_ptr = ((char *)ptr +
+                                      vaos->base.key.element[i].in.offset);
+         vaos->attrib[i].input_stride = stride;
       }
    }
 }
 
 
-static void vaos_destroy( struct draw_vs_varient *varient )
-{
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
-
-   if (vaos->machine)
-      align_free( vaos->machine );
-
-   x86_release_func( &vaos->func[0] );
-   x86_release_func( &vaos->func[1] );
-
-   FREE(vaos);
-}
 
 static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
                                       const unsigned *elts,
@@ -2127,6 +2153,10 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
    vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
+   vaos->machine->constants = vaos->draw->pt.user.constants;
+   vaos->machine->immediates = vaos->base.vs->immediates;
+   vaos->machine->attrib = vaos->attrib;
+
    vaos->gen_run_elts( varient,
                        elts,
                        count,
@@ -2141,6 +2171,10 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
    vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
+   vaos->machine->constants = vaos->draw->pt.user.constants;
+   vaos->machine->immediates = vaos->base.vs->immediates;
+   vaos->machine->attrib = vaos->attrib;
+
    vaos->gen_run_linear( varient,
                          start,
                          count,
@@ -2153,10 +2187,6 @@ static void vaos_set_constants( struct draw_vs_varient *varient,
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
-   memcpy(vaos->machine->constant,
-          constants,
-          (vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1) * 4 * sizeof(float));
-
 #if 0
    unsigned i;
    for (i =0; i < vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1; i++)
@@ -2187,6 +2217,21 @@ static void vaos_set_viewport( struct draw_vs_varient *varient,
    memcpy(vaos->machine->translate, viewport->translate, 4 * sizeof(float));
 }
 
+static void vaos_destroy( struct draw_vs_varient *varient )
+{
+   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+
+   if (vaos->machine)
+      align_free( vaos->machine );
+
+   FREE( vaos->attrib );
+
+   x86_release_func( &vaos->func[0] );
+   x86_release_func( &vaos->func[1] );
+
+   FREE(vaos);
+}
+
 
 
 static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
@@ -2207,6 +2252,11 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    vaos->base.run_elts = vaos_run_elts;
 
    vaos->draw = vs->draw;
+
+   vaos->attrib = MALLOC( key->nr_inputs * sizeof(vaos->attrib[0]) );
+   if (!vaos->attrib)
+      goto fail;
+
    vaos->machine = align_malloc( sizeof(struct aos_machine), 16 );
    if (!vaos->machine)
       goto fail;
@@ -2233,7 +2283,10 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    return &vaos->base;
 
  fail:
-   if (vaos->machine)
+   if (vaos && vaos->attrib)
+      FREE(vaos->attrib);
+
+   if (vaos && vaos->machine)
       align_free( vaos->machine );
 
    if (vaos)
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 837b32794f..295d2cb3fe 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -78,6 +78,14 @@ struct lit_info {
 #define MAX_SHINE_TAB    4
 #define MAX_LIT_INFO     16
 
+struct aos_attrib {
+   const void *input_ptr;
+   unsigned input_stride;
+};
+
+
+
+
 /* This is the temporary storage used by all the aos_sse vs varients.
  * Create one per context and reuse by passing a pointer in at
  * vs_varient creation??
@@ -86,8 +94,6 @@ struct aos_machine {
    float input    [MAX_INPUTS    ][4];
    float output   [MAX_OUTPUTS   ][4];
    float temp     [MAX_TEMPS     ][4];
-   float constant [MAX_CONSTANTS ][4]; /* fixme -- should just be a pointer */
-   float immediate[MAX_IMMEDIATES][4]; /* fixme -- should just be a pointer */
    float internal [MAX_INTERNALS ][4];
 
    float scale[4];              /* viewport */
@@ -105,12 +111,10 @@ struct aos_machine {
    ushort fpu_restore;
    ushort fpucntl;              /* one of FPU_* above */
 
-   struct {
-      const void *input_ptr;
-      unsigned input_stride;
+   const float (*immediates)[4];     /* points to shader data */
+   const float (*constants)[4];      /* points to draw data */
 
-      unsigned output_offset;
-   } attrib[PIPE_MAX_ATTRIBS];
+   const struct aos_attrib *attrib; /* points to ? */
 };
 
 
@@ -132,6 +136,7 @@ struct aos_compilation {
       unsigned last_used;
    } xmm[8];
 
+   unsigned ebp;                /* one of X86_* */
 
    boolean input_fetched[PIPE_MAX_ATTRIBS];
    unsigned output_last_write[PIPE_MAX_ATTRIBS];
@@ -148,6 +153,7 @@ struct aos_compilation {
    struct x86_reg outbuf_ECX;
    struct x86_reg machine_EDX;
    struct x86_reg count_ESI;    /* decrements to zero */
+   struct x86_reg temp_EBP;
 };
 
 struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp );
@@ -192,20 +198,20 @@ do {                                                                    \
 } while (0)
 
 
+#define X86_NULL       0
+#define X86_IMMEDIATES 1
+#define X86_CONSTANTS  2
+#define X86_ATTRIBS    3
 
-
+struct x86_reg aos_get_x86( struct aos_compilation *cp,
+                            unsigned value );
 
 
 struct draw_vs_varient_aos_sse {
    struct draw_vs_varient base;
    struct draw_context *draw;
 
-#if 0
-   struct {
-      const void *ptr;
-      unsigned stride;
-   } attrib[PIPE_MAX_ATTRIBS];
-#endif
+   struct aos_attrib *attrib;
 
    struct aos_machine *machine; /* XXX: temporarily unshared */
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index 836110f382..45e2092209 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -91,25 +91,25 @@ static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp,
 
 
 
-static void get_src_ptr( struct x86_function *func,
+static void get_src_ptr( struct aos_compilation *cp,
                          struct x86_reg src,
-                         struct x86_reg machine,
                          struct x86_reg elt,
                          unsigned a )
 {
-   struct x86_reg input_ptr = 
-      x86_make_disp(machine, 
-		    Offset(struct aos_machine, attrib[a].input_ptr));
+   struct x86_reg attrib = x86_make_disp(aos_get_x86( cp, X86_ATTRIBS ), 
+                                         a * sizeof(struct aos_attrib));
 
-   struct x86_reg input_stride = 
-      x86_make_disp(machine, 
-		    Offset(struct aos_machine, attrib[a].input_stride));
+   struct x86_reg input_ptr = x86_make_disp(attrib, 
+                                            Offset(struct aos_attrib, input_ptr));
+
+   struct x86_reg input_stride = x86_make_disp(attrib, 
+                                               Offset(struct aos_attrib, input_stride));
 
    /* Calculate pointer to current attrib:
     */
-   x86_mov(func, src, input_stride);
-   x86_imul(func, src, elt);
-   x86_add(func, src, input_ptr);
+   x86_mov(cp->func, src, input_stride);
+   x86_imul(cp->func, src, elt);
+   x86_add(cp->func, src, input_ptr);
 }
 
 
@@ -134,9 +134,8 @@ static boolean load_input( struct aos_compilation *cp,
 
    /* Figure out source pointer address:
     */
-   get_src_ptr(cp->func, 
+   get_src_ptr(cp, 
                src, 
-               cp->machine_EDX, 
                linear ? cp->idx_EBX : x86_deref(cp->idx_EBX),
                idx);
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 7781782ae8..24f619a278 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -68,8 +68,6 @@ struct draw_sse_vertex_shader {
    codegen_function func;
    
    struct tgsi_exec_machine *machine;
-
-   float immediates[TGSI_EXEC_NUM_IMMEDIATES][4];
 };
 
 
@@ -107,7 +105,7 @@ vs_sse_run_linear( struct draw_vertex_shader *base,
 		   machine->Outputs,
 		   (float (*)[4])constants,
 		   machine->Temps,
-		   shader->immediates,
+		   (float (*)[4])shader->base.immediates,
                    input,
                    base->info.num_inputs,
                    input_stride,
@@ -130,6 +128,8 @@ vs_sse_delete( struct draw_vertex_shader *base )
    
    x86_release_func( &shader->sse2_program );
 
+   align_free(shader->base.immediates);
+
    FREE( (void*) shader->base.state.tokens );
    FREE( shader );
 }
@@ -161,12 +161,18 @@ draw_create_vs_sse(struct draw_context *draw,
    vs->base.prepare = vs_sse_prepare;
    vs->base.run_linear = vs_sse_run_linear;
    vs->base.delete = vs_sse_delete;
+   
+   vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
+                                      sizeof(float), 16);
+
    vs->machine = &draw->vs.machine;
    
    x86_init_func( &vs->sse2_program );
 
    if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
-			&vs->sse2_program, vs->immediates, TRUE )) 
+			&vs->sse2_program, 
+                        (float (*)[4])vs->base.immediates, 
+                        TRUE )) 
       goto fail;
       
    vs->func = (codegen_function) x86_get_func( &vs->sse2_program );
-- 
cgit v1.2.3


From 62628c4d3d497cbca73fde869c9069fa90e6453e Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Thu, 29 May 2008 00:17:53 +0100
Subject: draw: share machine

---
 src/gallium/auxiliary/draw/Makefile                |   1 +
 src/gallium/auxiliary/draw/SConscript              |   1 +
 src/gallium/auxiliary/draw/draw_context.c          |   3 +
 src/gallium/auxiliary/draw/draw_private.h          |   9 +
 .../auxiliary/draw/draw_pt_fetch_shade_emit.c      |   6 -
 src/gallium/auxiliary/draw/draw_vs.c               |  43 ++-
 src/gallium/auxiliary/draw/draw_vs.h               |  24 +-
 src/gallium/auxiliary/draw/draw_vs_aos.c           | 282 ++-----------------
 src/gallium/auxiliary/draw/draw_vs_aos.h           |  25 +-
 src/gallium/auxiliary/draw/draw_vs_aos_machine.c   | 297 +++++++++++++++++++++
 src/gallium/auxiliary/draw/draw_vs_varient.c       |  49 +---
 11 files changed, 412 insertions(+), 328 deletions(-)
 create mode 100644 src/gallium/auxiliary/draw/draw_vs_aos_machine.c

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile
index 9a88ecc070..f2e36a89e9 100644
--- a/src/gallium/auxiliary/draw/Makefile
+++ b/src/gallium/auxiliary/draw/Makefile
@@ -37,6 +37,7 @@ C_SOURCES = \
 	draw_vs_varient.c \
 	draw_vs_aos.c \
 	draw_vs_aos_io.c \
+	draw_vs_aos_machine.c \
 	draw_vs_exec.c \
 	draw_vs_llvm.c \
 	draw_vs_sse.c 
diff --git a/src/gallium/auxiliary/draw/SConscript b/src/gallium/auxiliary/draw/SConscript
index 26919a2298..925e668f22 100644
--- a/src/gallium/auxiliary/draw/SConscript
+++ b/src/gallium/auxiliary/draw/SConscript
@@ -36,6 +36,7 @@ draw = env.ConvenienceLibrary(
 		'draw_vs.c',
 		'draw_vs_aos.c',
 		'draw_vs_aos_io.c',
+		'draw_vs_aos_machine.c',
 		'draw_vs_exec.c',
 		'draw_vs_llvm.c',
 		'draw_vs_sse.c',
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 2242074965..8509baf865 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -174,6 +174,8 @@ void draw_set_viewport_state( struct draw_context *draw,
                               viewport->translate[1] == 0.0f &&
                               viewport->translate[2] == 0.0f &&
                               viewport->translate[3] == 0.0f);
+
+   draw_vs_set_viewport( draw, viewport );
 }
 
 
@@ -218,6 +220,7 @@ draw_set_mapped_constant_buffer(struct draw_context *draw,
                                 const void *buffer)
 {
    draw->pt.user.constants = buffer;
+   draw_vs_set_constants( draw, (const float (*)[4])buffer );
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index c095bf3d7b..4cbccc8b5b 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -183,6 +183,10 @@ struct draw_context
        */
       struct gallivm_cpu_engine *engine;   
 
+      /* Here's another one:
+       */
+      struct aos_machine *aos_machine; 
+
 
       struct translate *fetch;
       struct translate_cache *fetch_cache;
@@ -215,6 +219,11 @@ struct draw_context
 boolean draw_vs_init( struct draw_context *draw );
 void draw_vs_destroy( struct draw_context *draw );
 
+void draw_vs_set_viewport( struct draw_context *, 
+                           const struct pipe_viewport_state * );
+
+void draw_vs_set_constants( struct draw_context *,
+                            const float (*constants)[4] );
 
 
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 729c7db999..5265a13160 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -189,12 +189,6 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
                               draw->pt.vertex_buffer[buf].pitch );
    }
 
-   fse->active->set_constants( fse->active,
-                               (const float (*)[4])draw->pt.user.constants );
-
-   fse->active->set_viewport( fse->active,
-                              &draw->viewport );
-
    //return TRUE;
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
index 9b899d404e..a8b6d0c90d 100644
--- a/src/gallium/auxiliary/draw/draw_vs.c
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -41,6 +41,22 @@
 
 
 
+
+void draw_vs_set_constants( struct draw_context *draw,
+                            const float (*constants)[4] )
+{
+   draw_vs_aos_machine_constants( draw->vs.aos_machine, constants );
+}
+
+
+void draw_vs_set_viewport( struct draw_context *draw,
+                           const struct pipe_viewport_state *viewport )
+{
+   draw_vs_aos_machine_viewport( draw->vs.aos_machine, viewport );
+}
+
+
+
 struct draw_vertex_shader *
 draw_create_vertex_shader(struct draw_context *draw,
                           const struct pipe_shader_state *shader)
@@ -83,6 +99,13 @@ void
 draw_delete_vertex_shader(struct draw_context *draw,
                           struct draw_vertex_shader *dvs)
 {
+   unsigned i;
+
+   for (i = 0; i < dvs->nr_varients; i++) 
+      dvs->varient[i]->destroy( dvs->varient[i] );
+
+   dvs->nr_varients = 0;
+
    dvs->delete( dvs );
 }
 
@@ -110,6 +133,10 @@ draw_vs_init( struct draw_context *draw )
    draw->vs.fetch_cache = translate_cache_create();
    if (!draw->vs.fetch_cache) 
       return FALSE;
+
+   draw->vs.aos_machine = draw_vs_aos_machine();
+   if (!draw->vs.aos_machine)
+      return FALSE;
       
    return TRUE;
 }
@@ -129,6 +156,9 @@ draw_vs_destroy( struct draw_context *draw )
    if (draw->vs.emit_cache)
       translate_cache_destroy(draw->vs.emit_cache);
 
+   if (draw->vs.aos_machine)
+      draw_vs_aos_machine_destroy(draw->vs.aos_machine);
+
    tgsi_exec_machine_free_data(&draw->vs.machine);
 
 }
@@ -153,10 +183,17 @@ draw_vs_lookup_varient( struct draw_vertex_shader *vs,
    if (varient == NULL)
       return NULL;
 
-   /* Add it to our list: 
+   /* Add it to our list, could be smarter: 
     */
-   assert(vs->nr_varients < Elements(vs->varient));
-   vs->varient[vs->nr_varients++] = varient;
+   if (vs->nr_varients < Elements(vs->varient)) {
+      vs->varient[vs->nr_varients++] = varient;
+   }
+   else {
+      vs->last_varient++;
+      vs->last_varient %= Elements(vs->varient);
+      vs->varient[vs->last_varient]->destroy(vs->varient[vs->last_varient]);
+      vs->varient[vs->last_varient] = varient;
+   }
 
    /* Done 
     */
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index 7aa0415baf..08c6de8ba8 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -70,16 +70,6 @@ struct draw_vs_varient_key {
 
 struct draw_vs_varient;
 
-typedef void (PIPE_CDECL *vsv_run_elts_func)( struct draw_vs_varient *,
-                                              const unsigned *elts,
-                                              unsigned count,
-                                              void *output_buffer);
-
-typedef void (PIPE_CDECL *vsv_run_linear_func)( struct draw_vs_varient *,
-                                                unsigned start,
-                                                unsigned count,
-                                                void *output_buffer);
-
 
 struct draw_vs_varient {
    struct draw_vs_varient_key key;
@@ -91,12 +81,6 @@ struct draw_vs_varient {
                       const void *ptr,
                       unsigned stride );
 
-   void (*set_constants)( struct draw_vs_varient *,
-                          const float (*constants)[4] );
-
-   void (*set_viewport)( struct draw_vs_varient *,
-                         const struct pipe_viewport_state * );
-
    void (PIPE_CDECL *run_linear)( struct draw_vs_varient *shader,
                                   unsigned start,
                                   unsigned count,
@@ -131,6 +115,7 @@ struct draw_vertex_shader {
     */
    struct draw_vs_varient *varient[16];
    unsigned nr_varients;
+   unsigned last_varient;
    struct draw_vs_varient *(*create_varient)( struct draw_vertex_shader *shader,
                                               const struct draw_vs_varient_key *key );
 
@@ -217,7 +202,14 @@ static INLINE int draw_vs_varient_key_compare( const struct draw_vs_varient_key
 }
 
 
+struct aos_machine *draw_vs_aos_machine( void );
+void draw_vs_aos_machine_destroy( struct aos_machine *machine );
+
+void draw_vs_aos_machine_constants( struct aos_machine *machine,
+                                    const float (*constants)[4] );
 
+void draw_vs_aos_machine_viewport( struct aos_machine *machine,
+                                   const struct pipe_viewport_state *viewport );
 
 
 #define MAX_TGSI_VERTICES 4
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 9056785e7a..b5e4e1e7b1 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -149,70 +149,7 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
 #define X87_CW_ROUND_MASK             (3<<10)
 #define X87_CW_INFINITY               (1<<12)
 
-static void do_populate_lut( struct shine_tab *tab,
-                             float unclamped_exponent )
-{
-   const float epsilon = 1.0F / 256.0F;    
-   float exponent = CLAMP(unclamped_exponent, -(128.0F - epsilon), (128.0F - epsilon));
-   unsigned i;
 
-   tab->exponent = unclamped_exponent; /* for later comparison */
-   
-   tab->values[0] = 0;
-   if (exponent == 0) {
-      for (i = 1; i < 258; i++) {
-         tab->values[i] = 1.0;
-      }      
-   }
-   else {
-      for (i = 1; i < 258; i++) {
-         tab->values[i] = powf((float)i * epsilon, exponent);
-      }
-   }
-}
-
-static void init_internals( struct aos_machine *machine )
-{
-   unsigned i;
-   float inv = 1.0f/255.0f;
-   float f255 = 255.0f;
-
-   ASSIGN_4V(machine->internal[IMM_SWZ],       1.0f,  -1.0f,  0.0f, 1.0f);
-   *(unsigned *)&machine->internal[IMM_SWZ][3] = 0xffffffff;
-
-   ASSIGN_4V(machine->internal[IMM_ONES],      1.0f,  1.0f,  1.0f,  1.0f);
-   ASSIGN_4V(machine->internal[IMM_NEGS],     -1.0f, -1.0f, -1.0f, -1.0f);
-   ASSIGN_4V(machine->internal[IMM_IDENTITY],  0.0f,  0.0f,  0.0f,  1.0f);
-   ASSIGN_4V(machine->internal[IMM_INV_255],   inv,   inv,   inv,   inv);
-   ASSIGN_4V(machine->internal[IMM_255],       f255,  f255,  f255,  f255);
-   ASSIGN_4V(machine->internal[IMM_RSQ],       -.5f,  1.5f,  0.0f,  0.0f);
-
-
-   machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP |
-                               X87_CW_EXCEPTION_DENORM_OP |
-                               X87_CW_EXCEPTION_ZERO_DIVIDE |
-                               X87_CW_EXCEPTION_OVERFLOW |
-                               X87_CW_EXCEPTION_UNDERFLOW |
-                               X87_CW_EXCEPTION_PRECISION |
-                               (1<<6) |
-                               X87_CW_ROUND_NEAREST |
-                               X87_CW_PRECISION_DOUBLE_EXT);
-
-   assert(machine->fpu_rnd_nearest == 0x37f);
-                               
-   machine->fpu_rnd_neg_inf = (X87_CW_EXCEPTION_INV_OP |
-                               X87_CW_EXCEPTION_DENORM_OP |
-                               X87_CW_EXCEPTION_ZERO_DIVIDE |
-                               X87_CW_EXCEPTION_OVERFLOW |
-                               X87_CW_EXCEPTION_UNDERFLOW |
-                               X87_CW_EXCEPTION_PRECISION |
-                               (1<<6) |
-                               X87_CW_ROUND_DOWN |
-                               X87_CW_PRECISION_DOUBLE_EXT);
-
-   for (i = 0; i < MAX_SHINE_TAB; i++)
-      do_populate_lut( &machine->shine_tab[i], 1.0f );
-}
 
 
 static void spill( struct aos_compilation *cp, unsigned idx )
@@ -1220,136 +1157,6 @@ static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_inst
    return TRUE;
 }
 
-static void PIPE_CDECL do_lit( struct aos_machine *machine,
-                               float *result,
-                               const float *in,
-                               unsigned count )
-{
-   if (in[0] > 0) 
-   {
-      if (in[1] <= 0.0) 
-      {
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = 1.0;
-         result[3] = 1.0F;
-      }
-      else
-      {
-         const float epsilon = 1.0F / 256.0F;    
-         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = powf(in[1], exponent);
-         result[3] = 1.0;
-      }
-   }
-   else 
-   {
-      result[0] = 1.0F;
-      result[1] = 0.0;
-      result[2] = 0.0;
-      result[3] = 1.0F;
-   }
-}
-
-
-static void PIPE_CDECL do_lit_lut( struct aos_machine *machine,
-                                   float *result,
-                                   const float *in,
-                                   unsigned count )
-{
-   if (in[0] > 0) 
-   {
-      if (in[1] <= 0.0) 
-      {
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = 1.0;
-         result[3] = 1.0F;
-         return;
-      }
-      
-      if (machine->lit_info[count].shine_tab->exponent != in[3]) {
-         machine->lit_info[count].func = do_lit;
-         goto no_luck;
-      }
-
-      if (in[1] <= 1.0)
-      {
-         const float *tab = machine->lit_info[count].shine_tab->values;
-         float f = in[1] * 256;
-         int k = (int)f;
-         float frac = f - (float)k;
-         
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = tab[k] + frac*(tab[k+1]-tab[k]);
-         result[3] = 1.0;
-         return;
-      }
-      
-   no_luck:
-      {
-         const float epsilon = 1.0F / 256.0F;    
-         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = powf(in[1], exponent);
-         result[3] = 1.0;
-      }
-   }
-   else 
-   {
-      result[0] = 1.0F;
-      result[1] = 0.0;
-      result[2] = 0.0;
-      result[3] = 1.0F;
-   }
-}
-
-
-
-static void PIPE_CDECL populate_lut( struct aos_machine *machine,
-                                     float *result,
-                                     const float *in,
-                                     unsigned count )
-{
-   unsigned i, tab;
-
-   /* Search for an existing table for this value.  Note that without
-    * static analysis we don't really know if in[3] will be constant,
-    * but it usually is...
-    */
-   for (tab = 0; tab < 4; tab++) {
-      if (machine->shine_tab[tab].exponent == in[3]) {
-         goto found;
-      }
-   }
-
-   for (tab = 0, i = 1; i < 4; i++) {
-      if (machine->shine_tab[i].last_used < machine->shine_tab[tab].last_used)
-         tab = i;
-   }
-
-   if (machine->shine_tab[tab].last_used == machine->now) {
-      /* No unused tables (this is not a ffvertex program...).  Just
-       * call pow each time:
-       */
-      machine->lit_info[count].func = do_lit;
-      machine->lit_info[count].func( machine, result, in, count );
-      return;
-   }
-   else {
-      do_populate_lut( &machine->shine_tab[tab], in[3] );
-   }
-
- found:
-   machine->shine_tab[tab].last_used = machine->now;
-   machine->lit_info[count].shine_tab = &machine->shine_tab[tab];
-   machine->lit_info[count].func = do_lit_lut;
-   machine->lit_info[count].func( machine, result, in, count );
-}
 
 
 
@@ -1413,7 +1220,7 @@ static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_inst
                                              Offset(struct lit_info, func)));
    }
    else {
-      x86_mov_reg_imm( cp->func, ecx, (int)do_lit );
+      x86_mov_reg_imm( cp->func, ecx, (int)aos_do_lit );
    }
 
    x86_call( cp->func, ecx );
@@ -1434,7 +1241,7 @@ static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_inst
    return TRUE;
 }
 
-   
+#if 0   
 static boolean emit_inline_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
 {
    struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); 
@@ -1495,6 +1302,7 @@ static boolean emit_inline_LIT( struct aos_compilation *cp, const struct tgsi_fu
 
    return TRUE;
 }
+#endif
 
 
 
@@ -1945,7 +1753,7 @@ static void find_last_write_outputs( struct aos_compilation *cp )
 }
 
 
-#define ARG_VARIENT    1
+#define ARG_MACHINE    1
 #define ARG_START_ELTS 2
 #define ARG_COUNT      3
 #define ARG_OUTBUF     4
@@ -1985,7 +1793,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
    /* Load arguments into regs:
     */
-   x86_mov(cp.func, cp.machine_EDX, x86_fn_arg(cp.func, ARG_VARIENT));
+   x86_mov(cp.func, cp.machine_EDX, x86_fn_arg(cp.func, ARG_MACHINE));
    x86_mov(cp.func, cp.idx_EBX, x86_fn_arg(cp.func, ARG_START_ELTS));
    x86_mov(cp.func, cp.count_ESI, x86_fn_arg(cp.func, ARG_COUNT));
    x86_mov(cp.func, cp.outbuf_ECX, x86_fn_arg(cp.func, ARG_OUTBUF));
@@ -1997,11 +1805,6 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    x86_cmp(cp.func, cp.count_ESI, cp.tmp_EAX);
    fixup = x86_jcc_forward(cp.func, cc_E);
 
-   /* Dig out the machine pointer from inside the varient arg 
-    */
-   x86_mov(cp.func, cp.machine_EDX, 
-           x86_make_disp(cp.machine_EDX,
-                         Offset( struct draw_vs_varient_aos_sse, machine )));
 
    save_fpu_state( &cp );
    set_fpu_round_nearest( &cp );
@@ -2151,13 +1954,14 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
                                       void *output_buffer )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct aos_machine *machine = vaos->draw->vs.aos_machine;
 
-   vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
-   vaos->machine->constants = vaos->draw->pt.user.constants;
-   vaos->machine->immediates = vaos->base.vs->immediates;
-   vaos->machine->attrib = vaos->attrib;
+   machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
+   machine->constants = (const float (*)[4])vaos->draw->pt.user.constants;
+   machine->immediates = vaos->base.vs->immediates;
+   machine->attrib = vaos->attrib;
 
-   vaos->gen_run_elts( varient,
+   vaos->gen_run_elts( machine,
                        elts,
                        count,
                        output_buffer );
@@ -2169,61 +1973,25 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
                                         void *output_buffer )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct aos_machine *machine = vaos->draw->vs.aos_machine;
 
-   vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
-   vaos->machine->constants = vaos->draw->pt.user.constants;
-   vaos->machine->immediates = vaos->base.vs->immediates;
-   vaos->machine->attrib = vaos->attrib;
+   machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
+   machine->constants = (const float (*)[4])vaos->draw->pt.user.constants;
+   machine->immediates = vaos->base.vs->immediates;
+   machine->attrib = vaos->attrib;
 
-   vaos->gen_run_linear( varient,
+   vaos->gen_run_linear( machine,
                          start,
                          count,
                          output_buffer );
 }
 
 
-static void vaos_set_constants( struct draw_vs_varient *varient,
-                                const float (*constants)[4] )
-{
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
-
-#if 0
-   unsigned i;
-   for (i =0; i < vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1; i++)
-      debug_printf("state %d: %f %f %f %f\n",
-                   i, 
-                   constants[i][0],
-                   constants[i][1],
-                   constants[i][2],
-                   constants[i][3]);
-#endif
-
-   {
-      unsigned i;
-      for (i = 0; i < MAX_LIT_INFO; i++) {
-         vaos->machine->lit_info[i].func = populate_lut;
-         vaos->machine->now++;
-      }
-   }
-}
-
-
-static void vaos_set_viewport( struct draw_vs_varient *varient,
-                               const struct pipe_viewport_state *viewport )
-{
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
-
-   memcpy(vaos->machine->scale, viewport->scale, 4 * sizeof(float));
-   memcpy(vaos->machine->translate, viewport->translate, 4 * sizeof(float));
-}
 
 static void vaos_destroy( struct draw_vs_varient *varient )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
-   if (vaos->machine)
-      align_free( vaos->machine );
-
    FREE( vaos->attrib );
 
    x86_release_func( &vaos->func[0] );
@@ -2245,8 +2013,6 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    vaos->base.key = *key;
    vaos->base.vs = vs;
    vaos->base.set_input = vaos_set_buffer;
-   vaos->base.set_constants = vaos_set_constants;
-   vaos->base.set_viewport = vaos_set_viewport;
    vaos->base.destroy = vaos_destroy;
    vaos->base.run_linear = vaos_run_linear;
    vaos->base.run_elts = vaos_run_elts;
@@ -2257,13 +2023,6 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    if (!vaos->attrib)
       goto fail;
 
-   vaos->machine = align_malloc( sizeof(struct aos_machine), 16 );
-   if (!vaos->machine)
-      goto fail;
-   
-   memset(vaos->machine, 0, sizeof(struct aos_machine));
-   init_internals(vaos->machine);
-
    tgsi_dump(vs->state.tokens, 0);
 
    if (!build_vertex_program( vaos, TRUE ))
@@ -2272,11 +2031,11 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    if (!build_vertex_program( vaos, FALSE ))
       goto fail;
 
-   vaos->gen_run_linear = (vsv_run_linear_func)x86_get_func(&vaos->func[0]);
+   vaos->gen_run_linear = (vaos_run_linear_func)x86_get_func(&vaos->func[0]);
    if (!vaos->gen_run_linear)
       goto fail;
 
-   vaos->gen_run_elts = (vsv_run_elts_func)x86_get_func(&vaos->func[1]);
+   vaos->gen_run_elts = (vaos_run_elts_func)x86_get_func(&vaos->func[1]);
    if (!vaos->gen_run_elts)
       goto fail;
 
@@ -2286,9 +2045,6 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    if (vaos && vaos->attrib)
       FREE(vaos->attrib);
 
-   if (vaos && vaos->machine)
-      align_free( vaos->machine );
-
    if (vaos)
       x86_release_func( &vaos->func[0] );
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 295d2cb3fe..89a9174151 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -60,10 +60,16 @@ struct x86_function;
 #define FPU_RND_NEAREST 2
 
 struct aos_machine;
-typedef void (PIPE_CDECL *lit_func)( struct aos_machine *,
+typedef void PIPE_CDECL (*lit_func)( struct aos_machine *,
                                     float *result,
                                     const float *in,
                                     unsigned count );
+
+PIPE_CDECL void aos_do_lit( struct aos_machine *machine,
+                            float *result,
+                            const float *in,
+                            unsigned count );
+
 struct shine_tab {
    float exponent;
    float values[258];
@@ -207,16 +213,25 @@ struct x86_reg aos_get_x86( struct aos_compilation *cp,
                             unsigned value );
 
 
+typedef void (PIPE_CDECL *vaos_run_elts_func)( struct aos_machine *,
+                                               const unsigned *elts,
+                                               unsigned count,
+                                               void *output_buffer);
+
+typedef void (PIPE_CDECL *vaos_run_linear_func)( struct aos_machine *,
+                                                unsigned start,
+                                                unsigned count,
+                                                void *output_buffer);
+
+
 struct draw_vs_varient_aos_sse {
    struct draw_vs_varient base;
    struct draw_context *draw;
 
    struct aos_attrib *attrib;
 
-   struct aos_machine *machine; /* XXX: temporarily unshared */
-
-   vsv_run_linear_func gen_run_linear;
-   vsv_run_elts_func gen_run_elts;
+   vaos_run_linear_func gen_run_linear;
+   vaos_run_elts_func gen_run_elts;
 
 
    struct x86_function func[2];
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
new file mode 100644
index 0000000000..53e999b191
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
@@ -0,0 +1,297 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/util/tgsi_parse.h"
+#include "tgsi/util/tgsi_util.h"
+#include "tgsi/exec/tgsi_exec.h"
+#include "draw_vs.h"
+#include "draw_vs_aos.h"
+#include "draw_vertex.h"
+
+#include "rtasm/rtasm_x86sse.h"
+
+
+#define X87_CW_EXCEPTION_INV_OP       (1<<0)
+#define X87_CW_EXCEPTION_DENORM_OP    (1<<1)
+#define X87_CW_EXCEPTION_ZERO_DIVIDE  (1<<2)
+#define X87_CW_EXCEPTION_OVERFLOW     (1<<3)
+#define X87_CW_EXCEPTION_UNDERFLOW    (1<<4)
+#define X87_CW_EXCEPTION_PRECISION    (1<<5)
+#define X87_CW_PRECISION_SINGLE       (0<<8)
+#define X87_CW_PRECISION_RESERVED     (1<<8)
+#define X87_CW_PRECISION_DOUBLE       (2<<8)
+#define X87_CW_PRECISION_DOUBLE_EXT   (3<<8)
+#define X87_CW_PRECISION_MASK         (3<<8)
+#define X87_CW_ROUND_NEAREST          (0<<10)
+#define X87_CW_ROUND_DOWN             (1<<10)
+#define X87_CW_ROUND_UP               (2<<10)
+#define X87_CW_ROUND_ZERO             (3<<10)
+#define X87_CW_ROUND_MASK             (3<<10)
+#define X87_CW_INFINITY               (1<<12)
+
+
+PIPE_CDECL void aos_do_lit( struct aos_machine *machine,
+                            float *result,
+                            const float *in,
+                            unsigned count )
+{
+   if (in[0] > 0) 
+   {
+      if (in[1] <= 0.0) 
+      {
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = 1.0;
+         result[3] = 1.0F;
+      }
+      else
+      {
+         const float epsilon = 1.0F / 256.0F;    
+         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = powf(in[1], exponent);
+         result[3] = 1.0;
+      }
+   }
+   else 
+   {
+      result[0] = 1.0F;
+      result[1] = 0.0;
+      result[2] = 0.0;
+      result[3] = 1.0F;
+   }
+}
+
+
+static PIPE_CDECL void do_lit_lut( struct aos_machine *machine,
+                                   float *result,
+                                   const float *in,
+                                   unsigned count )
+{
+   if (in[0] > 0) 
+   {
+      if (in[1] <= 0.0) 
+      {
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = 1.0;
+         result[3] = 1.0F;
+         return;
+      }
+      
+      if (machine->lit_info[count].shine_tab->exponent != in[3]) {
+         machine->lit_info[count].func = aos_do_lit;
+         goto no_luck;
+      }
+
+      if (in[1] <= 1.0)
+      {
+         const float *tab = machine->lit_info[count].shine_tab->values;
+         float f = in[1] * 256;
+         int k = (int)f;
+         float frac = f - (float)k;
+         
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = tab[k] + frac*(tab[k+1]-tab[k]);
+         result[3] = 1.0;
+         return;
+      }
+      
+   no_luck:
+      {
+         const float epsilon = 1.0F / 256.0F;    
+         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = powf(in[1], exponent);
+         result[3] = 1.0;
+      }
+   }
+   else 
+   {
+      result[0] = 1.0F;
+      result[1] = 0.0;
+      result[2] = 0.0;
+      result[3] = 1.0F;
+   }
+}
+
+
+static void do_populate_lut( struct shine_tab *tab,
+                             float unclamped_exponent )
+{
+   const float epsilon = 1.0F / 256.0F;    
+   float exponent = CLAMP(unclamped_exponent, -(128.0F - epsilon), (128.0F - epsilon));
+   unsigned i;
+
+   tab->exponent = unclamped_exponent; /* for later comparison */
+   
+   tab->values[0] = 0;
+   if (exponent == 0) {
+      for (i = 1; i < 258; i++) {
+         tab->values[i] = 1.0;
+      }      
+   }
+   else {
+      for (i = 1; i < 258; i++) {
+         tab->values[i] = powf((float)i * epsilon, exponent);
+      }
+   }
+}
+
+
+
+
+static void PIPE_CDECL populate_lut( struct aos_machine *machine,
+                                     float *result,
+                                     const float *in,
+                                     unsigned count )
+{
+   unsigned i, tab;
+
+   /* Search for an existing table for this value.  Note that without
+    * static analysis we don't really know if in[3] will be constant,
+    * but it usually is...
+    */
+   for (tab = 0; tab < 4; tab++) {
+      if (machine->shine_tab[tab].exponent == in[3]) {
+         goto found;
+      }
+   }
+
+   for (tab = 0, i = 1; i < 4; i++) {
+      if (machine->shine_tab[i].last_used < machine->shine_tab[tab].last_used)
+         tab = i;
+   }
+
+   if (machine->shine_tab[tab].last_used == machine->now) {
+      /* No unused tables (this is not a ffvertex program...).  Just
+       * call pow each time:
+       */
+      machine->lit_info[count].func = aos_do_lit;
+      machine->lit_info[count].func( machine, result, in, count );
+      return;
+   }
+   else {
+      do_populate_lut( &machine->shine_tab[tab], in[3] );
+   }
+
+ found:
+   machine->shine_tab[tab].last_used = machine->now;
+   machine->lit_info[count].shine_tab = &machine->shine_tab[tab];
+   machine->lit_info[count].func = do_lit_lut;
+   machine->lit_info[count].func( machine, result, in, count );
+}
+
+
+void draw_vs_aos_machine_constants( struct aos_machine *machine,
+                                    const float (*constants)[4] )
+{
+   machine->constants = constants;
+
+   {
+      unsigned i;
+      for (i = 0; i < MAX_LIT_INFO; i++) {
+         machine->lit_info[i].func = populate_lut;
+         machine->now++;
+      }
+   }
+}
+
+
+void draw_vs_aos_machine_viewport( struct aos_machine *machine,
+                                   const struct pipe_viewport_state *viewport )
+{
+   memcpy(machine->scale, viewport->scale, 4 * sizeof(float));
+   memcpy(machine->translate, viewport->translate, 4 * sizeof(float));
+}
+
+
+
+void draw_vs_aos_machine_destroy( struct aos_machine *machine )
+{
+   align_free(machine);
+}
+
+struct aos_machine *draw_vs_aos_machine( void )
+{
+   struct aos_machine *machine;
+   unsigned i;
+   float inv = 1.0f/255.0f;
+   float f255 = 255.0f;
+
+   machine = align_malloc(sizeof(struct aos_machine), 16);
+   if (!machine)
+      return NULL;
+
+   memset(machine, 0, sizeof(*machine));
+
+   ASSIGN_4V(machine->internal[IMM_SWZ],       1.0f,  -1.0f,  0.0f, 1.0f);
+   *(unsigned *)&machine->internal[IMM_SWZ][3] = 0xffffffff;
+
+   ASSIGN_4V(machine->internal[IMM_ONES],      1.0f,  1.0f,  1.0f,  1.0f);
+   ASSIGN_4V(machine->internal[IMM_NEGS],     -1.0f, -1.0f, -1.0f, -1.0f);
+   ASSIGN_4V(machine->internal[IMM_IDENTITY],  0.0f,  0.0f,  0.0f,  1.0f);
+   ASSIGN_4V(machine->internal[IMM_INV_255],   inv,   inv,   inv,   inv);
+   ASSIGN_4V(machine->internal[IMM_255],       f255,  f255,  f255,  f255);
+   ASSIGN_4V(machine->internal[IMM_RSQ],       -.5f,  1.5f,  0.0f,  0.0f);
+
+
+   machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP |
+                               X87_CW_EXCEPTION_DENORM_OP |
+                               X87_CW_EXCEPTION_ZERO_DIVIDE |
+                               X87_CW_EXCEPTION_OVERFLOW |
+                               X87_CW_EXCEPTION_UNDERFLOW |
+                               X87_CW_EXCEPTION_PRECISION |
+                               (1<<6) |
+                               X87_CW_ROUND_NEAREST |
+                               X87_CW_PRECISION_DOUBLE_EXT);
+
+   assert(machine->fpu_rnd_nearest == 0x37f);
+                               
+   machine->fpu_rnd_neg_inf = (X87_CW_EXCEPTION_INV_OP |
+                               X87_CW_EXCEPTION_DENORM_OP |
+                               X87_CW_EXCEPTION_ZERO_DIVIDE |
+                               X87_CW_EXCEPTION_OVERFLOW |
+                               X87_CW_EXCEPTION_UNDERFLOW |
+                               X87_CW_EXCEPTION_PRECISION |
+                               (1<<6) |
+                               X87_CW_ROUND_DOWN |
+                               X87_CW_PRECISION_DOUBLE_EXT);
+
+   for (i = 0; i < MAX_SHINE_TAB; i++)
+      do_populate_lut( &machine->shine_tab[i], 1.0f );
+
+   return machine;
+}
+
+
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 784ae41205..18cb06e374 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -44,8 +44,6 @@
 struct draw_vs_varient_generic {
    struct draw_vs_varient base;
 
-   struct pipe_viewport_state viewport;
-
    struct draw_vertex_shader *shader;
    struct draw_context *draw;
    
@@ -57,21 +55,11 @@ struct draw_vs_varient_generic {
     */
    struct translate *fetch;
    struct translate *emit;
-
-   const float (*constants)[4];
 };
 
 
 
 
-static void vsvg_set_constants( struct draw_vs_varient *varient,
-                                const float (*constants)[4] )
-{
-   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
-
-   vsvg->constants = constants;
-}
-
 
 static void vsvg_set_input( struct draw_vs_varient *varient,
                             unsigned buffer,
@@ -94,8 +82,8 @@ static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg,
                              void *output_buffer )
 {
    char *ptr = (char *)output_buffer;
-   const float *scale = vsvg->viewport.scale;
-   const float *trans = vsvg->viewport.translate;
+   const float *scale = vsvg->base.vs->draw->viewport.scale;
+   const float *trans = vsvg->base.vs->draw->viewport.translate;
    unsigned stride = vsvg->base.key.output_stride;
    unsigned j;
 
@@ -115,8 +103,8 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg,
                              void *output_buffer )
 {
    char *ptr = (char *)output_buffer;
-   const float *scale = vsvg->viewport.scale;
-   const float *trans = vsvg->viewport.translate;
+   const float *scale = vsvg->base.vs->draw->viewport.scale;
+   const float *trans = vsvg->base.vs->draw->viewport.translate;
    unsigned stride = vsvg->base.key.output_stride;
    unsigned j;
 
@@ -130,10 +118,10 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg,
 }
                          
 
-static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
-                                      const unsigned *elts,
-                                      unsigned count,
-                                      void *output_buffer )
+static void vsvg_run_elts( struct draw_vs_varient *varient,
+                           const unsigned *elts,
+                           unsigned count,
+                           void *output_buffer)
 {
    struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
 			
@@ -150,7 +138,7 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
       vsvg->base.vs->run_linear( vsvg->base.vs, 
                                  output_buffer,
                                  output_buffer,
-                                 vsvg->constants,
+                                 (const float (*)[4])vsvg->base.vs->draw->pt.user.constants,
                                  count,
                                  vsvg->base.key.output_stride, 
                                  vsvg->base.key.output_stride);
@@ -186,10 +174,10 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
 }
 
 
-static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
-                                        unsigned start,
-                                        unsigned count,
-                                        void *output_buffer )
+static void vsvg_run_linear( struct draw_vs_varient *varient,
+                                   unsigned start,
+                                   unsigned count,
+                                   void *output_buffer )
 {
    struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
 	
@@ -206,7 +194,7 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
       vsvg->base.vs->run_linear( vsvg->base.vs, 
                                  output_buffer,
                                  output_buffer,
-                                 vsvg->constants,
+                                 (const float (*)[4])vsvg->base.vs->draw->pt.user.constants,
                                  count,
                                  vsvg->base.key.output_stride, 
                                  vsvg->base.key.output_stride);
@@ -245,13 +233,6 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
 
 
 
-static void vsvg_set_viewport( struct draw_vs_varient *varient,
-                               const struct pipe_viewport_state *viewport )
-{
-   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
-
-   vsvg->viewport = *viewport;
-}
 
 static void vsvg_destroy( struct draw_vs_varient *varient )
 {
@@ -272,8 +253,6 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    vsvg->base.key = *key;
    vsvg->base.vs = vs;
    vsvg->base.set_input     = vsvg_set_input;
-   vsvg->base.set_constants = vsvg_set_constants;
-   vsvg->base.set_viewport  = vsvg_set_viewport;
    vsvg->base.run_elts      = vsvg_run_elts;
    vsvg->base.run_linear    = vsvg_run_linear;
    vsvg->base.destroy       = vsvg_destroy;
-- 
cgit v1.2.3


From 6945bcb89370501e0a218bc656e68e30e4dadcda Mon Sep 17 00:00:00 2001
From: José Fonseca <jrfonseca@tungstengraphics.com>
Date: Thu, 29 May 2008 18:28:02 +0900
Subject: draw: Put PIPE_CDECL in the right places.

MSVC seems picky about this.
---
 src/gallium/auxiliary/draw/draw_vs_aos.h         |  4 ++--
 src/gallium/auxiliary/draw/draw_vs_aos_machine.c |  4 ++--
 src/gallium/auxiliary/draw/draw_vs_varient.c     | 16 ++++++++--------
 3 files changed, 12 insertions(+), 12 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 89a9174151..665b425e68 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -60,12 +60,12 @@ struct x86_function;
 #define FPU_RND_NEAREST 2
 
 struct aos_machine;
-typedef void PIPE_CDECL (*lit_func)( struct aos_machine *,
+typedef void (PIPE_CDECL *lit_func)( struct aos_machine *,
                                     float *result,
                                     const float *in,
                                     unsigned count );
 
-PIPE_CDECL void aos_do_lit( struct aos_machine *machine,
+void PIPE_CDECL aos_do_lit( struct aos_machine *machine,
                             float *result,
                             const float *in,
                             unsigned count );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
index 53e999b191..b7864e9f2f 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
@@ -57,7 +57,7 @@
 #define X87_CW_INFINITY               (1<<12)
 
 
-PIPE_CDECL void aos_do_lit( struct aos_machine *machine,
+void PIPE_CDECL aos_do_lit( struct aos_machine *machine,
                             float *result,
                             const float *in,
                             unsigned count )
@@ -91,7 +91,7 @@ PIPE_CDECL void aos_do_lit( struct aos_machine *machine,
 }
 
 
-static PIPE_CDECL void do_lit_lut( struct aos_machine *machine,
+static void PIPE_CDECL do_lit_lut( struct aos_machine *machine,
                                    float *result,
                                    const float *in,
                                    unsigned count )
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 18cb06e374..4a155251b5 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -118,10 +118,10 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg,
 }
                          
 
-static void vsvg_run_elts( struct draw_vs_varient *varient,
-                           const unsigned *elts,
-                           unsigned count,
-                           void *output_buffer)
+static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
+                                      const unsigned *elts,
+                                      unsigned count,
+                                      void *output_buffer)
 {
    struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
 			
@@ -174,10 +174,10 @@ static void vsvg_run_elts( struct draw_vs_varient *varient,
 }
 
 
-static void vsvg_run_linear( struct draw_vs_varient *varient,
-                                   unsigned start,
-                                   unsigned count,
-                                   void *output_buffer )
+static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
+                                        unsigned start,
+                                        unsigned count,
+                                        void *output_buffer )
 {
    struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
 	
-- 
cgit v1.2.3


From 8f887b4252208e60e7e86217ec3b72fb639a4e82 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Thu, 29 May 2008 13:26:01 +0100
Subject: draw: michal's patch for calling powf...  teapot still not quite
 right

---
 src/gallium/auxiliary/draw/draw_vs_aos.c | 46 ++++++++++++++++++++++++++++++++
 src/gallium/auxiliary/draw/draw_vs_aos.h |  1 +
 2 files changed, 47 insertions(+)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 55cabb6df9..725f36b502 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -1372,11 +1372,20 @@ static boolean emit_MAD( struct aos_compilation *cp, const struct tgsi_full_inst
    return TRUE;
 }
 
+/* A wrapper for powf().
+ * Makes sure it is cdecl and operates on floats.
+ */
+static float PIPE_CDECL _powerf( float x, float y )
+{
+   return powf( x, y );
+}
+
 /* Really not sufficient -- need to check for conditions that could
  * generate inf/nan values, which will slow things down hugely.
  */
 static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
 {
+#if 0
    x87_fld_src(cp, &op->FullSrcRegisters[1], 0);  /* a1.x */
    x87_fld_src(cp, &op->FullSrcRegisters[0], 0);	/* a0.x a1.x */
    x87_fyl2x(cp->func);	                                /* a1*log2(a0) */
@@ -1384,6 +1393,42 @@ static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_inst
    x87_emit_ex2( cp );		/* 2^(a1*log2(a0)) */
 
    x87_fstp_dest4(cp, &op->FullDstRegisters[0]);
+#else
+   uint i;
+
+   /* For absolute correctness, need to spill/invalidate all XMM regs
+    * too.  
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+
+   /* Push caller-save (ie scratch) regs.  
+    */
+   x86_cdecl_caller_push_regs( cp->func );
+
+   x86_lea( cp->func, cp->stack_ESP, x86_make_disp(cp->stack_ESP, -8) );
+
+   x87_fld_src( cp, &op->FullSrcRegisters[1], 0 );
+   x87_fstp( cp->func, x86_make_disp( cp->stack_ESP, 4 ) );
+   x87_fld_src( cp, &op->FullSrcRegisters[0], 0 );
+   x87_fstp( cp->func, x86_make_disp( cp->stack_ESP, 0 ) );
+
+   x86_mov_reg_imm( cp->func, cp->tmp_EAX, (unsigned long) _powerf );
+   x86_call( cp->func, cp->tmp_EAX );
+
+   x86_lea( cp->func, cp->stack_ESP, x86_make_disp(cp->stack_ESP, 8) );
+
+   x86_cdecl_caller_pop_regs( cp->func );
+
+   /* Note retval on x87 stack:
+    */
+   cp->func->x87_stack++;
+
+   x87_fstp_dest4( cp, &op->FullDstRegisters[0] );
+#endif
    return TRUE;
 }
 
@@ -1781,6 +1826,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    cp.machine_EDX   = x86_make_reg(file_REG32, reg_DX);
    cp.count_ESI     = x86_make_reg(file_REG32, reg_SI);
    cp.temp_EBP     = x86_make_reg(file_REG32, reg_BP);
+   cp.stack_ESP = x86_make_reg( file_REG32, reg_SP );
 
    x86_init_func(cp.func);
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 665b425e68..fb6d43d32e 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -160,6 +160,7 @@ struct aos_compilation {
    struct x86_reg machine_EDX;
    struct x86_reg count_ESI;    /* decrements to zero */
    struct x86_reg temp_EBP;
+   struct x86_reg stack_ESP;
 };
 
 struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp );
-- 
cgit v1.2.3


From f27c7729a98937a761eacceabdfd03f9d694d257 Mon Sep 17 00:00:00 2001
From: José Fonseca <jrfonseca@tungstengraphics.com>
Date: Fri, 6 Jun 2008 13:29:59 +0900
Subject: draw: Compile draw_vs_aos only on x86.

---
 src/gallium/auxiliary/draw/draw_vs_aos.h         | 4 ++++
 src/gallium/auxiliary/draw/draw_vs_aos_machine.c | 6 ++++++
 2 files changed, 10 insertions(+)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index fb6d43d32e..f6f6af2dd9 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -31,6 +31,9 @@
 #ifndef DRAW_VS_AOS_H
 #define DRAW_VS_AOS_H
 
+#include "pipe/p_config.h"
+
+#ifdef PIPE_ARCH_X86
 
 struct tgsi_token;
 struct x86_function;
@@ -239,6 +242,7 @@ struct draw_vs_varient_aos_sse {
 };
 
 
+#endif
 
 #endif 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
index b7864e9f2f..dcb6c2c8d4 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
@@ -26,6 +26,10 @@
  **************************************************************************/
 
 
+#include "pipe/p_config.h"
+
+#ifdef PIPE_ARCH_X86
+
 #include "pipe/p_util.h"
 #include "pipe/p_shader_tokens.h"
 #include "tgsi/util/tgsi_parse.h"
@@ -295,3 +299,5 @@ struct aos_machine *draw_vs_aos_machine( void )
 }
 
 
+#endif
+
-- 
cgit v1.2.3


From bd9264210097d08073a4ea3619ca25db56245280 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Wed, 11 Jun 2008 09:36:00 +0100
Subject: draw: remove another debug assert on failover to generic vs varient

---
 src/gallium/auxiliary/draw/draw_vs_aos.h | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index f6f6af2dd9..66944a4e33 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -202,9 +202,8 @@ struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp,
 
 #define ERROR(cp, msg)                                                  \
 do {                                                                    \
-   debug_printf("%s: x86 translation failed: %s\n", __FUNCTION__, msg); \
+   if (0) debug_printf("%s: x86 translation failed: %s\n", __FUNCTION__, msg); \
    cp->error = 1;                                                       \
-   assert(0);                                                           \
 } while (0)
 
 
-- 
cgit v1.2.3


From 08f1b8ac709105d42ec34f8b8a81421e3b0fbc81 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Thu, 12 Jun 2008 16:01:05 -0600
Subject: gallium: fix SSE codegen for instructions that use both a CONSTANT
 and IMMEDIATE

Fixes codegen for instructions like MUL dst, CONST[0], IMM[0];  the two operands
would up getting aliased in the x86/sse code.

Fixes glean/vertProg1/fogparams test.
---
 src/gallium/auxiliary/draw/draw_vs_aos.c    | 26 +++++++++++++++++++-------
 src/gallium/auxiliary/draw/draw_vs_aos.h    |  3 ++-
 src/gallium/auxiliary/draw/draw_vs_aos_io.c |  2 +-
 3 files changed, 22 insertions(+), 9 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 5d4a8b38c8..388dd3fbee 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -67,19 +67,30 @@ static INLINE boolean eq( struct x86_reg a,
 }
       
 struct x86_reg aos_get_x86( struct aos_compilation *cp,
+                            unsigned which_reg, /* quick hack */
                             unsigned value )
 {
-   if (cp->ebp != value) {
+   struct x86_reg reg;
+
+   if (which_reg == 0)
+      reg = cp->temp_EBP;
+   else
+      reg = cp->tmp_EAX;
+
+   if (cp->x86_reg[which_reg] != value) {
       unsigned offset;
 
       switch (value) {
       case X86_IMMEDIATES:
+         assert(which_reg == 0);
          offset = Offset(struct aos_machine, immediates);
          break;
       case X86_CONSTANTS:
+         assert(which_reg == 1);
          offset = Offset(struct aos_machine, constants);
          break;
       case X86_ATTRIBS:
+         assert(which_reg == 0);
          offset = Offset(struct aos_machine, attrib);
          break;
       default:
@@ -87,14 +98,14 @@ struct x86_reg aos_get_x86( struct aos_compilation *cp,
          offset = 0;
       }
 
-      x86_mov(cp->func, cp->temp_EBP, 
+
+      x86_mov(cp->func, reg, 
               x86_make_disp(cp->machine_EDX, offset));
-      /* x86_deref(x86_make_disp(cp->machine_EDX, offset))); */
 
-      cp->ebp = value;
+      cp->x86_reg[which_reg] = value;
    }
 
-   return cp->temp_EBP;
+   return reg;
 }
 
 
@@ -118,10 +129,10 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
       return x86_make_disp(ptr, Offset(struct aos_machine, internal[idx]));
 
    case TGSI_FILE_IMMEDIATE: 
-      return x86_make_disp(aos_get_x86(cp, X86_IMMEDIATES), idx * 4 * sizeof(float));
+      return x86_make_disp(aos_get_x86(cp, 0, X86_IMMEDIATES), idx * 4 * sizeof(float));
 
    case TGSI_FILE_CONSTANT: 
-      return x86_make_disp(aos_get_x86(cp, X86_CONSTANTS), idx * 4 * sizeof(float));
+      return x86_make_disp(aos_get_x86(cp, 1, X86_CONSTANTS), idx * 4 * sizeof(float));
 
    default:
       ERROR(cp, "unknown reg file");
@@ -1413,6 +1424,7 @@ static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_inst
    x87_fld_src( cp, &op->FullSrcRegisters[0], 0 );
    x87_fstp( cp->func, x86_make_disp( cp->stack_ESP, 0 ) );
 
+   /* tmp_EAX has been pushed & will be restored below */
    x86_mov_reg_imm( cp->func, cp->tmp_EAX, (unsigned long) _powerf );
    x86_call( cp->func, cp->tmp_EAX );
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 66944a4e33..64e021ff6b 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -145,7 +145,7 @@ struct aos_compilation {
       unsigned last_used;
    } xmm[8];
 
-   unsigned ebp;                /* one of X86_* */
+   unsigned x86_reg[2];                /* one of X86_* */
 
    boolean input_fetched[PIPE_MAX_ATTRIBS];
    unsigned output_last_write[PIPE_MAX_ATTRIBS];
@@ -213,6 +213,7 @@ do {                                                                    \
 #define X86_ATTRIBS    3
 
 struct x86_reg aos_get_x86( struct aos_compilation *cp,
+                            unsigned which_reg,
                             unsigned value );
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index b720185709..6b92811870 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -96,7 +96,7 @@ static void get_src_ptr( struct aos_compilation *cp,
                          struct x86_reg elt,
                          unsigned a )
 {
-   struct x86_reg attrib = x86_make_disp(aos_get_x86( cp, X86_ATTRIBS ), 
+   struct x86_reg attrib = x86_make_disp(aos_get_x86( cp, 0, X86_ATTRIBS ), 
                                          a * sizeof(struct aos_attrib));
 
    struct x86_reg input_ptr = x86_make_disp(attrib, 
-- 
cgit v1.2.3


From feea0c9d958bc1645b09b288cd4d4756d0d6e61a Mon Sep 17 00:00:00 2001
From: Brian Paul <brian.paul@tungstengraphics.com>
Date: Tue, 2 Sep 2008 18:04:08 -0600
Subject: gallium: fix out of bounds array errors in SSE execution

1. #define MAX_INPUTS/OUTPUTS/TEMPS/etc with better values.
2. Add assertions in aos_get_x86() to check register file indexes
3. Assert that constant regs haven't changed after running SSE code.
---
 src/gallium/auxiliary/draw/draw_vs_aos.c | 11 +++++++++++
 src/gallium/auxiliary/draw/draw_vs_aos.h | 10 +++++-----
 2 files changed, 16 insertions(+), 5 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 760fcb389f..a556477a76 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -32,6 +32,7 @@
 #include "util/u_memory.h"
 #include "util/u_math.h"
 #include "pipe/p_shader_tokens.h"
+#include "pipe/p_debug.h"
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_util.h"
 #include "tgsi/tgsi_exec.h"
@@ -119,21 +120,26 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
 
    switch (file) {
    case TGSI_FILE_INPUT:
+      assert(idx < MAX_INPUTS);
       return x86_make_disp(ptr, Offset(struct aos_machine, input[idx]));
 
    case TGSI_FILE_OUTPUT:
       return x86_make_disp(ptr, Offset(struct aos_machine, output[idx]));
 
    case TGSI_FILE_TEMPORARY:
+      assert(idx < MAX_TEMPS);
       return x86_make_disp(ptr, Offset(struct aos_machine, temp[idx]));
 
    case AOS_FILE_INTERNAL:
+      assert(idx < MAX_INTERNALS);
       return x86_make_disp(ptr, Offset(struct aos_machine, internal[idx]));
 
    case TGSI_FILE_IMMEDIATE: 
+      assert(idx < MAX_IMMEDIATES);  /* just a sanity check */
       return x86_make_disp(aos_get_x86(cp, 0, X86_IMMEDIATES), idx * 4 * sizeof(float));
 
    case TGSI_FILE_CONSTANT: 
+      assert(idx < MAX_CONSTANTS);  /* just a sanity check */
       return x86_make_disp(aos_get_x86(cp, 1, X86_CONSTANTS), idx * 4 * sizeof(float));
 
    default:
@@ -2108,6 +2114,11 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
                          start,
                          count,
                          output_buffer );
+
+   /* Sanity spot checks to make sure we didn't trash our constants */
+   assert(machine->internal[IMM_ONES][0] == 1.0f);
+   assert(machine->internal[IMM_IDENTITY][0] == 0.0f);
+   assert(machine->internal[IMM_NEGS][0] == -1.0f);
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 64e021ff6b..7fe6f79db0 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -51,11 +51,11 @@ struct x86_function;
 #define W    3
 
 #define MAX_INPUTS     PIPE_MAX_ATTRIBS
-#define MAX_OUTPUTS    PIPE_MAX_ATTRIBS
-#define MAX_TEMPS      PIPE_MAX_ATTRIBS /* say */
-#define MAX_CONSTANTS  PIPE_MAX_ATTRIBS /* say */
-#define MAX_IMMEDIATES PIPE_MAX_ATTRIBS /* say */
-#define MAX_INTERNALS  8
+#define MAX_OUTPUTS    PIPE_MAX_SHADER_OUTPUTS
+#define MAX_TEMPS      TGSI_EXEC_NUM_TEMPS
+#define MAX_CONSTANTS  1024  /** only used for sanity checking */
+#define MAX_IMMEDIATES 1024  /** only used for sanity checking */
+#define MAX_INTERNALS  8     /** see IMM_x values below */
 
 #define AOS_FILE_INTERNAL TGSI_FILE_COUNT
 
-- 
cgit v1.2.3


From a15699c3f54edb5d5b42960e7568e587b752e407 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Wed, 1 Oct 2008 13:34:38 +0100
Subject: draw: add streamlined paths for fetching linear verts

---
 src/gallium/auxiliary/draw/draw_vs_aos.c    |  44 +++++----
 src/gallium/auxiliary/draw/draw_vs_aos.h    |  19 ++--
 src/gallium/auxiliary/draw/draw_vs_aos_io.c | 137 +++++++++++++++++++++-------
 3 files changed, 134 insertions(+), 66 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index a556477a76..4c794e0e23 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -92,9 +92,9 @@ struct x86_reg aos_get_x86( struct aos_compilation *cp,
          assert(which_reg == 1);
          offset = Offset(struct aos_machine, constants);
          break;
-      case X86_ATTRIBS:
+      case X86_BUFFERS:
          assert(which_reg == 0);
-         offset = Offset(struct aos_machine, attrib);
+         offset = Offset(struct aos_machine, buffer);
          break;
       default:
          assert(0);
@@ -1939,6 +1939,8 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    save_fpu_state( &cp );
    set_fpu_round_nearest( &cp );
 
+   aos_init_inputs( &cp, linear );
+
    /* Note address for loop jump 
     */
    label = x86_get_label(cp.func);
@@ -2018,13 +2020,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
       /* Incr index
        */   
-      if (linear) {
-         x86_inc(cp.func, cp.idx_EBX);
-      } 
-      else {
-         x86_lea(cp.func, cp.idx_EBX, x86_make_disp(cp.idx_EBX, 4));
-      }
-
+      aos_incr_inputs( &cp, linear );
    }
    /* decr count, loop if not zero
     */
@@ -2065,14 +2061,10 @@ static void vaos_set_buffer( struct draw_vs_varient *varient,
                              unsigned stride )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
-   unsigned i;
 
-   for (i = 0; i < vaos->base.key.nr_inputs; i++) {
-      if (vaos->base.key.element[i].in.buffer == buf) {
-         vaos->attrib[i].input_ptr = ((char *)ptr +
-                                      vaos->base.key.element[i].in.offset);
-         vaos->attrib[i].input_stride = stride;
-      }
+   if (buf < vaos->nr_vb) {
+      vaos->buffer[buf].base_ptr = (char *)ptr;
+      vaos->buffer[buf].stride = stride;
    }
 }
 
@@ -2089,7 +2081,7 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
    machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
    machine->constants = vaos->draw->vs.aligned_constants;
    machine->immediates = vaos->base.vs->immediates;
-   machine->attrib = vaos->attrib;
+   machine->buffer = vaos->buffer;
 
    vaos->gen_run_elts( machine,
                        elts,
@@ -2108,7 +2100,7 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
    machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
    machine->constants = vaos->draw->vs.aligned_constants;
    machine->immediates = vaos->base.vs->immediates;
-   machine->attrib = vaos->attrib;
+   machine->buffer = vaos->buffer;
 
    vaos->gen_run_linear( machine,
                          start,
@@ -2127,7 +2119,7 @@ static void vaos_destroy( struct draw_vs_varient *varient )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
-   FREE( vaos->attrib );
+   FREE( vaos->buffer );
 
    x86_release_func( &vaos->func[0] );
    x86_release_func( &vaos->func[1] );
@@ -2140,6 +2132,7 @@ static void vaos_destroy( struct draw_vs_varient *varient )
 static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
                                                  const struct draw_vs_varient_key *key )
 {
+   unsigned i;
    struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse);
 
    if (!vaos)
@@ -2154,10 +2147,15 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
 
    vaos->draw = vs->draw;
 
-   vaos->attrib = MALLOC( key->nr_inputs * sizeof(vaos->attrib[0]) );
-   if (!vaos->attrib)
+   for (i = 0; i < key->nr_inputs; i++) 
+      vaos->nr_vb = MAX2( vaos->nr_vb, key->element[i].in.buffer + 1 );
+
+   vaos->buffer = MALLOC( vaos->nr_vb * sizeof(vaos->buffer[0]) );
+   if (!vaos->buffer)
       goto fail;
 
+   debug_printf("nr_vb: %d\n", vaos->nr_vb);
+
 #if 0
    tgsi_dump(vs->state.tokens, 0);
 #endif
@@ -2179,8 +2177,8 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    return &vaos->base;
 
  fail:
-   if (vaos && vaos->attrib)
-      FREE(vaos->attrib);
+   if (vaos && vaos->buffer)
+      FREE(vaos->buffer);
 
    if (vaos)
       x86_release_func( &vaos->func[0] );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 7fe6f79db0..306392e5d6 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -87,9 +87,10 @@ struct lit_info {
 #define MAX_SHINE_TAB    4
 #define MAX_LIT_INFO     16
 
-struct aos_attrib {
-   const void *input_ptr;
-   unsigned input_stride;
+struct aos_buffer {
+   const void *base_ptr;
+   unsigned stride;
+   void *ptr;                   /* updated per vertex */
 };
 
 
@@ -123,7 +124,7 @@ struct aos_machine {
    const float (*immediates)[4];     /* points to shader data */
    const float (*constants)[4];      /* points to draw data */
 
-   const struct aos_attrib *attrib; /* points to ? */
+   const struct aos_buffer *buffer; /* points to ? */
 };
 
 
@@ -179,8 +180,9 @@ struct x86_reg aos_get_shader_reg( struct aos_compilation *cp,
                                    unsigned file,
                                    unsigned idx );
 
-boolean aos_fetch_inputs( struct aos_compilation *cp,
-                          boolean linear );
+boolean aos_init_inputs( struct aos_compilation *cp, boolean linear );
+boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear );
+boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear );
 
 boolean aos_emit_outputs( struct aos_compilation *cp );
 
@@ -210,7 +212,7 @@ do {                                                                    \
 #define X86_NULL       0
 #define X86_IMMEDIATES 1
 #define X86_CONSTANTS  2
-#define X86_ATTRIBS    3
+#define X86_BUFFERS    3
 
 struct x86_reg aos_get_x86( struct aos_compilation *cp,
                             unsigned which_reg,
@@ -232,7 +234,8 @@ struct draw_vs_varient_aos_sse {
    struct draw_vs_varient base;
    struct draw_context *draw;
 
-   struct aos_attrib *attrib;
+   struct aos_buffer *buffer;
+   unsigned nr_vb;
 
    vaos_run_linear_func gen_run_linear;
    vaos_run_elts_func gen_run_elts;
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index 26297c74f8..8e08b9285f 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -95,28 +95,6 @@ static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp,
 
 
 
-static void get_src_ptr( struct aos_compilation *cp,
-                         struct x86_reg src,
-                         struct x86_reg elt,
-                         unsigned a )
-{
-   struct x86_reg attrib = x86_make_disp(aos_get_x86( cp, 0, X86_ATTRIBS ), 
-                                         a * sizeof(struct aos_attrib));
-
-   struct x86_reg input_ptr = x86_make_disp(attrib, 
-                                            Offset(struct aos_attrib, input_ptr));
-
-   struct x86_reg input_stride = x86_make_disp(attrib, 
-                                               Offset(struct aos_attrib, input_stride));
-
-   /* Calculate pointer to current attrib:
-    */
-   x86_mov(cp->func, src, input_stride);
-   x86_imul(cp->func, src, elt);
-   x86_add(cp->func, src, input_ptr);
-}
-
-
 /* Extended swizzles?  Maybe later.
  */  
 static void emit_swizzle( struct aos_compilation *cp,
@@ -128,22 +106,44 @@ static void emit_swizzle( struct aos_compilation *cp,
 }
 
 
+
+static boolean get_buffer_ptr( struct aos_compilation *cp,
+                            unsigned buf_idx,
+                            struct x86_reg elt,
+                            struct x86_reg ptr)
+{
+   struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                      buf_idx * sizeof(struct aos_buffer));
+
+   struct x86_reg buf_base_ptr = x86_make_disp(buf, 
+                                               Offset(struct aos_buffer, base_ptr));
+
+   struct x86_reg buf_stride = x86_make_disp(buf, 
+                                             Offset(struct aos_buffer, stride));
+
+   /* Calculate pointer to current attrib:
+    */
+   x86_mov(cp->func, ptr, buf_stride);
+   x86_imul(cp->func, ptr, elt);
+   x86_add(cp->func, ptr, buf_base_ptr);
+
+   return TRUE;
+}
+
+
+
+
 static boolean load_input( struct aos_compilation *cp,
                            unsigned idx,
-                           boolean linear )
+                           struct x86_reg bufptr )
 {
    unsigned format = cp->vaos->base.key.element[idx].in.format;
-   struct x86_reg src = cp->tmp_EAX;
+   unsigned offset = cp->vaos->base.key.element[idx].in.offset;
    struct x86_reg dataXMM = aos_get_xmm_reg(cp);
 
    /* Figure out source pointer address:
     */
-   get_src_ptr(cp, 
-               src, 
-               linear ? cp->idx_EBX : x86_deref(cp->idx_EBX),
-               idx);
-
-   src = x86_deref(src);
+   struct x86_reg src = x86_make_disp(bufptr, offset);
 
    aos_adopt_xmm_reg( cp,
                       dataXMM,
@@ -179,20 +179,87 @@ static boolean load_input( struct aos_compilation *cp,
    return TRUE;
 }
 
-
-boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear )
+static boolean load_inputs( struct aos_compilation *cp,
+                            unsigned buffer,
+                            struct x86_reg ptr )
 {
    unsigned i;
-   
+
    for (i = 0; i < cp->vaos->base.key.nr_inputs; i++) {
-      if (!load_input( cp, i, linear ))
+      if (cp->vaos->base.key.element[i].in.buffer == buffer) {
+
+         if (!load_input( cp, i, ptr ))
+            return FALSE;
+
+         cp->insn_counter++;
+      }
+   }
+   
+   return TRUE;
+}
+
+boolean aos_init_inputs( struct aos_compilation *cp, boolean linear )
+{
+   if (linear && cp->vaos->nr_vb == 1) {
+
+      struct x86_reg elt = cp->idx_EBX;
+      struct x86_reg ptr = cp->tmp_EAX;
+
+      if (!get_buffer_ptr( cp, 0, elt, ptr ))
          return FALSE;
-      cp->insn_counter++;
+
+      /* In the linear, single buffer case, keep the buffer pointer
+       * instead of the index number.
+       */
+      x86_mov( cp->func, elt, ptr );
+   }
+
+   return TRUE;
+}
+
+boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear )
+{
+   if (linear && cp->vaos->nr_vb == 1) {
+      
+      load_inputs( cp, 0, cp->idx_EBX );
+
+   }
+   else {
+      struct x86_reg elt = linear ? cp->idx_EBX : x86_deref(cp->idx_EBX);
+      unsigned j;
+   
+      for (j = 0; j < cp->vaos->nr_vb; j++) {
+         struct x86_reg ptr = cp->tmp_EAX;
+
+         if (!get_buffer_ptr( cp, j, elt, ptr ))
+            return FALSE;
+
+         cp->insn_counter++;
+
+         if (!load_inputs( cp, j, ptr ))
+            return FALSE;
+      }
    }
 
    return TRUE;
 }
 
+boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear )
+{
+   if (linear && cp->vaos->nr_vb == 1) {
+      struct x86_reg stride = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                            (0 * sizeof(struct aos_buffer) + 
+                                             Offset(struct aos_buffer, stride)));
+
+      x86_add(cp->func, cp->idx_EBX, stride);
+   }
+   else if (linear) {
+      x86_inc(cp->func, cp->idx_EBX);
+   } 
+   else {
+      x86_lea(cp->func, cp->idx_EBX, x86_make_disp(cp->idx_EBX, 4));
+   }
+}
 
 
 
-- 
cgit v1.2.3


From 21f98ad30aaeab5085d12278830f485e61b47cc1 Mon Sep 17 00:00:00 2001
From: Keith Whitwell <keith@tungstengraphics.com>
Date: Wed, 1 Oct 2008 18:40:01 +0100
Subject: draw: don't keep refetching constant inputs

---
 .../auxiliary/draw/draw_pt_fetch_shade_emit.c      |  37 +++---
 src/gallium/auxiliary/draw/draw_vs.h               |   4 +-
 src/gallium/auxiliary/draw/draw_vs_aos.c           |  26 ++++-
 src/gallium/auxiliary/draw/draw_vs_aos.h           |   2 +
 src/gallium/auxiliary/draw/draw_vs_aos_io.c        | 127 +++++++++++++++------
 src/gallium/auxiliary/draw/draw_vs_varient.c       |  10 +-
 6 files changed, 144 insertions(+), 62 deletions(-)

(limited to 'src/gallium/auxiliary/draw/draw_vs_aos.h')

diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 73fc70c1bc..a0e08dd10a 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -79,6 +79,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
    unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
    const struct vertex_info *vinfo;
    unsigned i;
+   unsigned nr_vbs = 0;
    
 
    if (!draw->render->set_primitive( draw->render, 
@@ -102,7 +103,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
 
    fse->key.viewport = !draw->identity_viewport;
    fse->key.clip = !draw->bypass_clipping;
-   fse->key.pad = 0;
+   fse->key.const_vbuffers = 0;
 
    memset(fse->key.element, 0, 
           fse->key.nr_elements * sizeof(fse->key.element[0]));
@@ -116,9 +117,16 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
        */
       fse->key.element[i].in.buffer = src->vertex_buffer_index;
       fse->key.element[i].in.offset = src->src_offset;
+      nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1);
    }
    
+   for (i = 0; i < 5 && i < nr_vbs; i++) {
+      if (draw->pt.vertex_buffer[i].pitch == 0)
+         fse->key.const_vbuffers |= (1<<i);
+   }
 
+   if (0) debug_printf("%s: lookup const_vbuffers: %x\n", __FUNCTION__, fse->key.const_vbuffers);
+   
    {
       unsigned dst_offset = 0;
 
@@ -162,13 +170,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
       }
    }
 
-
-   /* Would normally look up a vertex shader and peruse its list of
-    * varients somehow.  We omitted that step and put all the
-    * hardcoded "shaders" into an array.  We're just making the
-    * assumption that this happens to be a matching shader...  ie
-    * you're running isosurf, aren't you?
-    */
+   
    fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader, 
                                          &fse->key );
 
@@ -177,18 +179,17 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
       return ;
    }
 
+   if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__, 
+                       fse->active->key.const_vbuffers);
+
    /* Now set buffer pointers:
     */
-   for (i = 0; i < num_vs_inputs; i++) {
-      unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index;
-
-      fse->active->set_input( fse->active, 
-                              i, 
-                              
-                              ((const ubyte *) draw->pt.user.vbuffer[buf] + 
-                               draw->pt.vertex_buffer[buf].buffer_offset),
-                              
-                              draw->pt.vertex_buffer[buf].pitch );
+   for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
+      fse->active->set_buffer( fse->active, 
+                               i, 
+                               ((const ubyte *) draw->pt.user.vbuffer[i] + 
+                                draw->pt.vertex_buffer[i].buffer_offset),
+                              draw->pt.vertex_buffer[i].pitch );
    }
 
    *max_vertices = (draw->render->max_vertex_buffer_bytes / 
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index 45992d1986..68c24abad3 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -64,7 +64,7 @@ struct draw_vs_varient_key {
    unsigned nr_outputs:8;
    unsigned viewport:1;
    unsigned clip:1;
-   unsigned pad:5;
+   unsigned const_vbuffers:5;
    struct draw_varient_element element[PIPE_MAX_ATTRIBS];
 };
 
@@ -76,7 +76,7 @@ struct draw_vs_varient {
 
    struct draw_vertex_shader *vs;
 
-   void (*set_input)( struct draw_vs_varient *,
+   void (*set_buffer)( struct draw_vs_varient *,
                       unsigned i,
                       const void *ptr,
                       unsigned stride );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 4c794e0e23..87232865e2 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -196,6 +196,18 @@ static void spill( struct aos_compilation *cp, unsigned idx )
 }
 
 
+void aos_spill_all( struct aos_compilation *cp )
+{
+   unsigned i;
+
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+}
+
+
 static struct x86_reg get_xmm_writable( struct aos_compilation *cp,
                                         struct x86_reg reg )
 {
@@ -1941,6 +1953,9 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
    aos_init_inputs( &cp, linear );
 
+   cp.x86_reg[0] = 0;
+   cp.x86_reg[1] = 0;
+   
    /* Note address for loop jump 
     */
    label = x86_get_label(cp.func);
@@ -2066,6 +2081,8 @@ static void vaos_set_buffer( struct draw_vs_varient *varient,
       vaos->buffer[buf].base_ptr = (char *)ptr;
       vaos->buffer[buf].stride = stride;
    }
+
+   if (0) debug_printf("%s %d/%d: %p %d\n", __FUNCTION__, buf, vaos->nr_vb, ptr, stride);
 }
 
 
@@ -2078,6 +2095,8 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
 
+   if (0) debug_printf("%s %d\n", __FUNCTION__, count);
+
    machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
    machine->constants = vaos->draw->vs.aligned_constants;
    machine->immediates = vaos->base.vs->immediates;
@@ -2097,6 +2116,9 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
 
+   if (0) debug_printf("%s %d %d const: %x\n", __FUNCTION__, start, count, 
+                       vaos->base.key.const_vbuffers);
+
    machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
    machine->constants = vaos->draw->vs.aligned_constants;
    machine->immediates = vaos->base.vs->immediates;
@@ -2140,7 +2162,7 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    
    vaos->base.key = *key;
    vaos->base.vs = vs;
-   vaos->base.set_input = vaos_set_buffer;
+   vaos->base.set_buffer = vaos_set_buffer;
    vaos->base.destroy = vaos_destroy;
    vaos->base.run_linear = vaos_run_linear;
    vaos->base.run_elts = vaos_run_elts;
@@ -2154,7 +2176,7 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    if (!vaos->buffer)
       goto fail;
 
-   debug_printf("nr_vb: %d\n", vaos->nr_vb);
+   debug_printf("nr_vb: %d const: %x\n", vaos->nr_vb, vaos->base.key.const_vbuffers);
 
 #if 0
    tgsi_dump(vs->state.tokens, 0);
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 306392e5d6..264387517b 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -176,6 +176,8 @@ void aos_adopt_xmm_reg( struct aos_compilation *cp,
                         unsigned idx,
                         unsigned dirty );
 
+void aos_spill_all( struct aos_compilation *cp );
+
 struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, 
                                    unsigned file,
                                    unsigned idx );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index 8e08b9285f..b0c51d7fa1 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -108,29 +108,45 @@ static void emit_swizzle( struct aos_compilation *cp,
 
 
 static boolean get_buffer_ptr( struct aos_compilation *cp,
-                            unsigned buf_idx,
-                            struct x86_reg elt,
-                            struct x86_reg ptr)
+                               boolean linear,
+                               unsigned buf_idx,
+                               struct x86_reg elt,
+                               struct x86_reg ptr)
 {
    struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
                                       buf_idx * sizeof(struct aos_buffer));
 
-   struct x86_reg buf_base_ptr = x86_make_disp(buf, 
-                                               Offset(struct aos_buffer, base_ptr));
-
    struct x86_reg buf_stride = x86_make_disp(buf, 
                                              Offset(struct aos_buffer, stride));
+   if (linear) {
+      struct x86_reg buf_ptr = x86_make_disp(buf, 
+                                             Offset(struct aos_buffer, ptr));
 
-   /* Calculate pointer to current attrib:
-    */
-   x86_mov(cp->func, ptr, buf_stride);
-   x86_imul(cp->func, ptr, elt);
-   x86_add(cp->func, ptr, buf_base_ptr);
 
-   return TRUE;
-}
+      /* Calculate pointer to current attrib:
+       */
+      x86_mov(cp->func, ptr, buf_ptr);
+      x86_mov(cp->func, elt, buf_stride);
+      x86_add(cp->func, elt, ptr);
+      sse_prefetchnta(cp->func, x86_deref(elt));
+      x86_mov(cp->func, buf_ptr, elt);
+   }
+   else {
+      struct x86_reg buf_base_ptr = x86_make_disp(buf, 
+                                                  Offset(struct aos_buffer, base_ptr));
+
+
+      /* Calculate pointer to current attrib:
+       */
+      x86_mov(cp->func, ptr, buf_stride);
+      x86_imul(cp->func, ptr, elt);
+      x86_add(cp->func, ptr, buf_base_ptr);
+   }
 
+   cp->insn_counter++;
 
+   return TRUE;
+}
 
 
 static boolean load_input( struct aos_compilation *cp,
@@ -200,18 +216,57 @@ static boolean load_inputs( struct aos_compilation *cp,
 
 boolean aos_init_inputs( struct aos_compilation *cp, boolean linear )
 {
-   if (linear && cp->vaos->nr_vb == 1) {
+   unsigned i;
+   for (i = 0; i < cp->vaos->nr_vb; i++) {
+      struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                         i * sizeof(struct aos_buffer));
 
-      struct x86_reg elt = cp->idx_EBX;
-      struct x86_reg ptr = cp->tmp_EAX;
+      struct x86_reg buf_base_ptr = x86_make_disp(buf, 
+                                                  Offset(struct aos_buffer, base_ptr));
 
-      if (!get_buffer_ptr( cp, 0, elt, ptr ))
-         return FALSE;
+      if (cp->vaos->base.key.const_vbuffers & (1<<i)) {
+         struct x86_reg ptr = cp->tmp_EAX;
 
-      /* In the linear, single buffer case, keep the buffer pointer
-       * instead of the index number.
-       */
-      x86_mov( cp->func, elt, ptr );
+         x86_mov(cp->func, ptr, buf_base_ptr);
+
+         /* Load all inputs for this constant vertex buffer
+          */
+         load_inputs( cp, i, x86_deref(ptr) );
+         
+         /* Then just force them out to aos_machine.input[]
+          */
+         aos_spill_all( cp );
+
+      }
+      else if (linear) {
+
+         struct x86_reg elt = cp->idx_EBX;
+         struct x86_reg ptr = cp->tmp_EAX;
+
+         struct x86_reg buf_stride = x86_make_disp(buf, 
+                                                   Offset(struct aos_buffer, stride));
+
+         struct x86_reg buf_ptr = x86_make_disp(buf, 
+                                                Offset(struct aos_buffer, ptr));
+
+
+         /* Calculate pointer to current attrib:
+          */
+         x86_mov(cp->func, ptr, buf_stride);
+         x86_imul(cp->func, ptr, elt);
+         x86_add(cp->func, ptr, buf_base_ptr);
+
+
+         /* In the linear case, keep the buffer pointer instead of the
+          * index number.
+          */
+         if (cp->vaos->nr_vb == 1) 
+            x86_mov( cp->func, elt, ptr );
+         else
+            x86_mov( cp->func, buf_ptr, ptr );
+
+         cp->insn_counter++;
+      }
    }
 
    return TRUE;
@@ -219,23 +274,22 @@ boolean aos_init_inputs( struct aos_compilation *cp, boolean linear )
 
 boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear )
 {
-   if (linear && cp->vaos->nr_vb == 1) {
-      
-      load_inputs( cp, 0, cp->idx_EBX );
+   unsigned j;
 
-   }
-   else {
-      struct x86_reg elt = linear ? cp->idx_EBX : x86_deref(cp->idx_EBX);
-      unsigned j;
-   
-      for (j = 0; j < cp->vaos->nr_vb; j++) {
+   for (j = 0; j < cp->vaos->nr_vb; j++) {
+      if (cp->vaos->base.key.const_vbuffers & (1<<j)) {
+         /* just retreive pre-transformed input */
+      }
+      else if (linear && cp->vaos->nr_vb == 1) {
+         load_inputs( cp, 0, cp->idx_EBX );
+      }
+      else {
+         struct x86_reg elt = linear ? cp->idx_EBX : x86_deref(cp->idx_EBX);
          struct x86_reg ptr = cp->tmp_EAX;
 
-         if (!get_buffer_ptr( cp, j, elt, ptr ))
+         if (!get_buffer_ptr( cp, linear, j, elt, ptr ))
             return FALSE;
 
-         cp->insn_counter++;
-
          if (!load_inputs( cp, j, ptr ))
             return FALSE;
       }
@@ -252,13 +306,16 @@ boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear )
                                              Offset(struct aos_buffer, stride)));
 
       x86_add(cp->func, cp->idx_EBX, stride);
+      sse_prefetchnta(cp->func, x86_deref(cp->idx_EBX));
    }
    else if (linear) {
-      x86_inc(cp->func, cp->idx_EBX);
+      /* Nothing to do */
    } 
    else {
       x86_lea(cp->func, cp->idx_EBX, x86_make_disp(cp->idx_EBX, 4));
    }
+
+   return TRUE;
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 4daf05dae7..7ee567d478 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -64,10 +64,10 @@ struct draw_vs_varient_generic {
 
 
 
-static void vsvg_set_input( struct draw_vs_varient *varient,
-                            unsigned buffer,
-                            const void *ptr,
-                            unsigned stride )
+static void vsvg_set_buffer( struct draw_vs_varient *varient,
+                             unsigned buffer,
+                             const void *ptr,
+                             unsigned stride )
 {
    struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
 
@@ -265,7 +265,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
 
    vsvg->base.key = *key;
    vsvg->base.vs = vs;
-   vsvg->base.set_input     = vsvg_set_input;
+   vsvg->base.set_buffer    = vsvg_set_buffer;
    vsvg->base.run_elts      = vsvg_run_elts;
    vsvg->base.run_linear    = vsvg_run_linear;
    vsvg->base.destroy       = vsvg_destroy;
-- 
cgit v1.2.3