i965: Move repeat-instruction-suppression to batchbuffer core

Move the tracking of the last emitted instructions into the core
batchbuffer routines and take advantage of the shadow batch copy to
avoid extra memory allocations and copies.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

6 files changed, 46 insertions, 137 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_cc.c b/src/mesa/drivers/dri/i965/brw_cc.c
index c37376ef0d..b7048ecb39 100644
--- a/src/mesa/drivers/dri/i965/brw_cc.c
+++ b/src/mesa/drivers/dri/i965/brw_cc.c
@@ -233,18 +233,16 @@ const struct brw_tracked_state brw_cc_unit = {
 
 static void upload_blend_constant_color(struct brw_context *brw)
 {
-   struct gl_context *ctx = &brw->intel.ctx;
-   struct brw_blend_constant_color bcc;
-
-   memset(&bcc, 0, sizeof(bcc));
-   bcc.header.opcode = _3DSTATE_BLEND_CONSTANT_COLOR;
-   bcc.header.length = sizeof(bcc)/4-2;
-   bcc.blend_constant_color[0] = ctx->Color.BlendColor[0];
-   bcc.blend_constant_color[1] = ctx->Color.BlendColor[1];
-   bcc.blend_constant_color[2] = ctx->Color.BlendColor[2];
-   bcc.blend_constant_color[3] = ctx->Color.BlendColor[3];
-
-   BRW_CACHED_BATCH_STRUCT(brw, &bcc);
+   struct intel_context *intel = &brw->intel;
+   struct gl_context *ctx = &intel->ctx;
+
+   BEGIN_BATCH(5);
+   OUT_BATCH(_3DSTATE_BLEND_CONSTANT_COLOR << 16 | (5-2));
+   OUT_BATCH(ctx->Color.BlendColor[0]);
+   OUT_BATCH(ctx->Color.BlendColor[1]);
+   OUT_BATCH(ctx->Color.BlendColor[2]);
+   OUT_BATCH(ctx->Color.BlendColor[3]);
+   CACHED_BATCH();
 }
 
 const struct brw_tracked_state brw_blend_constant_color = {
diff --git a/src/mesa/drivers/dri/i965/brw_curbe.c b/src/mesa/drivers/dri/i965/brw_curbe.c
index 877b22fec1..ae11c487a2 100644
--- a/src/mesa/drivers/dri/i965/brw_curbe.c
+++ b/src/mesa/drivers/dri/i965/brw_curbe.c
@@ -146,22 +146,24 @@ const struct brw_tracked_state brw_curbe_offsets = {
  */
 void brw_upload_cs_urb_state(struct brw_context *brw)
 {
-   struct brw_cs_urb_state cs_urb;
-   memset(&cs_urb, 0, sizeof(cs_urb));
+   struct intel_context *intel = &brw->intel;
 
+   BEGIN_BATCH(2);
    /* It appears that this is the state packet for the CS unit, ie. the
     * urb entries detailed here are housed in the CS range from the
     * URB_FENCE command.
     */
-   cs_urb.header.opcode = CMD_CS_URB_STATE;
-   cs_urb.header.length = sizeof(cs_urb)/4 - 2;
+   OUT_BATCH(CMD_CS_URB_STATE << 16 | (2-2));
 
    /* BRW_NEW_URB_FENCE */
-   cs_urb.bits0.nr_urb_entries = brw->urb.nr_cs_entries;
-   cs_urb.bits0.urb_entry_size = brw->urb.csize - 1;
-
-   assert(brw->urb.nr_cs_entries);
-   BRW_CACHED_BATCH_STRUCT(brw, &cs_urb);
+   if (brw->urb.csize == 0) {
+      OUT_BATCH(0);
+   } else {
+      /* BRW_NEW_URB_FENCE */
+      assert(brw->urb.nr_cs_entries);
+      OUT_BATCH((brw->urb.csize - 1) << 4 | brw->urb.nr_cs_entries);
+   }
+   CACHED_BATCH();
 }
 
 static GLfloat fixed_plane[6][4] = {
diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c
index c5c7409824..c768be23fa 100644
--- a/src/mesa/drivers/dri/i965/brw_misc_state.c
+++ b/src/mesa/drivers/dri/i965/brw_misc_state.c
@@ -301,16 +301,15 @@ const struct brw_tracked_state brw_depthbuffer = {
 
 static void upload_polygon_stipple(struct brw_context *brw)
 {
+   struct intel_context *intel = &brw->intel;
    struct gl_context *ctx = &brw->intel.ctx;
-   struct brw_polygon_stipple bps;
    GLuint i;
 
    if (!ctx->Polygon.StippleFlag)
       return;
 
-   memset(&bps, 0, sizeof(bps));
-   bps.header.opcode = _3DSTATE_POLY_STIPPLE_PATTERN;
-   bps.header.length = sizeof(bps)/4-2;
+   BEGIN_BATCH(33);
+   OUT_BATCH(_3DSTATE_POLY_STIPPLE_PATTERN << 16 | (33 - 2));
 
    /* Polygon stipple is provided in OpenGL order, i.e. bottom
     * row first.  If we're rendering to a window (i.e. the
@@ -321,14 +320,13 @@ static void upload_polygon_stipple(struct brw_context *brw)
     */
    if (ctx->DrawBuffer->Name == 0) {
       for (i = 0; i < 32; i++)
-         bps.stipple[i] = ctx->PolygonStipple[31 - i]; /* invert */
+	  OUT_BATCH(ctx->PolygonStipple[31 - i]); /* invert */
    }
    else {
       for (i = 0; i < 32; i++)
-         bps.stipple[i] = ctx->PolygonStipple[i]; /* don't invert */
+	 OUT_BATCH(ctx->PolygonStipple[i]);
    }
-
-   BRW_CACHED_BATCH_STRUCT(brw, &bps);
+   CACHED_BATCH();
 }
 
 const struct brw_tracked_state brw_polygon_stipple = {
@@ -347,15 +345,14 @@ const struct brw_tracked_state brw_polygon_stipple = {
 
 static void upload_polygon_stipple_offset(struct brw_context *brw)
 {
+   struct intel_context *intel = &brw->intel;
    struct gl_context *ctx = &brw->intel.ctx;
-   struct brw_polygon_stipple_offset bpso;
 
    if (!ctx->Polygon.StippleFlag)
       return;
 
-   memset(&bpso, 0, sizeof(bpso));
-   bpso.header.opcode = _3DSTATE_POLY_STIPPLE_OFFSET;
-   bpso.header.length = sizeof(bpso)/4-2;
+   BEGIN_BATCH(2);
+   OUT_BATCH(_3DSTATE_POLY_STIPPLE_OFFSET << 16 | (2-2));
 
    /* If we're drawing to a system window (ctx->DrawBuffer->Name == 0),
     * we have to invert the Y axis in order to match the OpenGL
@@ -365,16 +362,11 @@ static void upload_polygon_stipple_offset(struct brw_context *brw)
     * system works just fine, and there's no window system to
     * worry about.
     */
-   if (brw->intel.ctx.DrawBuffer->Name == 0) {
-      bpso.bits0.x_offset = 0;
-      bpso.bits0.y_offset = (32 - (ctx->DrawBuffer->Height & 31)) & 31;
-   }
-   else {
-      bpso.bits0.y_offset = 0;
-      bpso.bits0.x_offset = 0;
-   }
-
-   BRW_CACHED_BATCH_STRUCT(brw, &bpso);
+   if (brw->intel.ctx.DrawBuffer->Name == 0)
+      OUT_BATCH((32 - (ctx->DrawBuffer->Height & 31)) & 31);
+   else
+      OUT_BATCH(0);
+   CACHED_BATCH();
 }
 
 #define _NEW_WINDOW_POS 0x40000000
@@ -393,18 +385,17 @@ const struct brw_tracked_state brw_polygon_stipple_offset = {
  */
 static void upload_aa_line_parameters(struct brw_context *brw)
 {
+   struct intel_context *intel = &brw->intel;
    struct gl_context *ctx = &brw->intel.ctx;
-   struct brw_aa_line_parameters balp;
 
    if (!ctx->Line.SmoothFlag || !brw->has_aa_line_parameters)
       return;
 
+   OUT_BATCH(_3DSTATE_AA_LINE_PARAMETERS << 16 | (3 - 2));
    /* use legacy aa line coverage computation */
-   memset(&balp, 0, sizeof(balp));
-   balp.header.opcode = _3DSTATE_AA_LINE_PARAMETERS;
-   balp.header.length = sizeof(balp) / 4 - 2;
-   
-   BRW_CACHED_BATCH_STRUCT(brw, &balp);
+   OUT_BATCH(0);
+   OUT_BATCH(0);
+   CACHED_BATCH();
 }
 
 const struct brw_tracked_state brw_aa_line_parameters = {
@@ -422,28 +413,21 @@ const struct brw_tracked_state brw_aa_line_parameters = {
 
 static void upload_line_stipple(struct brw_context *brw)
 {
+   struct intel_context *intel = &brw->intel;
    struct gl_context *ctx = &brw->intel.ctx;
-   struct brw_line_stipple bls;
    GLfloat tmp;
    GLint tmpi;
 
    if (!ctx->Line.StippleFlag)
       return;
 
-   memset(&bls, 0, sizeof(bls));
-   bls.header.opcode = _3DSTATE_LINE_STIPPLE_PATTERN;
-   bls.header.length = sizeof(bls)/4 - 2;
-
-   bls.bits0.pattern = ctx->Line.StipplePattern;
-   bls.bits1.repeat_count = ctx->Line.StippleFactor;
-
+   BEGIN_BATCH(3);
+   OUT_BATCH(_3DSTATE_LINE_STIPPLE_PATTERN << 16 | (3 - 2));
+   OUT_BATCH(ctx->Line.StipplePattern);
    tmp = 1.0 / (GLfloat) ctx->Line.StippleFactor;
    tmpi = tmp * (1<<13);
-
-
-   bls.bits1.inverse_repeat_count = tmpi;
-
-   BRW_CACHED_BATCH_STRUCT(brw, &bls);
+   OUT_BATCH(tmpi << 16 | ctx->Line.StippleFactor);
+   CACHED_BATCH();
 }
 
 const struct brw_tracked_state brw_line_stipple = {
diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h
index 23d9e90c6e..86b0caa4a4 100644
--- a/src/mesa/drivers/dri/i965/brw_state.h
+++ b/src/mesa/drivers/dri/i965/brw_state.h
@@ -166,13 +166,7 @@ void brw_destroy_caches( struct brw_context *brw );
  */
 #define BRW_BATCH_STRUCT(brw, s) intel_batchbuffer_data(&brw->intel, (s), \
 							sizeof(*(s)), false)
-#define BRW_CACHED_BATCH_STRUCT(brw, s) brw_cached_batch_struct( brw, (s), sizeof(*(s)) )
 
-GLboolean brw_cached_batch_struct( struct brw_context *brw,
-				   const void *data,
-				   GLuint sz );
-void brw_destroy_batch_cache( struct brw_context *brw );
-void brw_clear_batch_cache( struct brw_context *brw );
 void *brw_state_batch(struct brw_context *brw,
 		      int size,
 		      int alignment,
diff --git a/src/mesa/drivers/dri/i965/brw_state_batch.c b/src/mesa/drivers/dri/i965/brw_state_batch.c
index f363a922c1..213c7a38d8 100644
--- a/src/mesa/drivers/dri/i965/brw_state_batch.c
+++ b/src/mesa/drivers/dri/i965/brw_state_batch.c
@@ -29,75 +29,10 @@
   *   Keith Whitwell <keith@tungstengraphics.com>
   */
      
-
-
 #include "brw_state.h"
 #include "intel_batchbuffer.h"
 #include "main/imports.h"
 
-
-
-/* A facility similar to the data caching code above, which aims to
- * prevent identical commands being issued repeatedly.
- */
-GLboolean brw_cached_batch_struct( struct brw_context *brw,
-				   const void *data,
-				   GLuint sz )
-{
-   struct brw_cached_batch_item *item = brw->cached_batch_items;
-   struct header *newheader = (struct header *)data;
-
-   if (brw->emit_state_always) {
-      intel_batchbuffer_data(&brw->intel, data, sz, false);
-      return GL_TRUE;
-   }
-
-   while (item) {
-      if (item->header->opcode == newheader->opcode) {
-	 if (item->sz == sz && memcmp(item->header, newheader, sz) == 0)
-	    return GL_FALSE;
-	 if (item->sz != sz) {
-	    free(item->header);
-	    item->header = malloc(sz);
-	    item->sz = sz;
-	 }
-	 goto emit;
-      }
-      item = item->next;
-   }
-
-   assert(!item);
-   item = CALLOC_STRUCT(brw_cached_batch_item);
-   item->header = malloc(sz);
-   item->sz = sz;
-   item->next = brw->cached_batch_items;
-   brw->cached_batch_items = item;
-
- emit:
-   memcpy(item->header, newheader, sz);
-   intel_batchbuffer_data(&brw->intel, data, sz, false);
-   return GL_TRUE;
-}
-
-void brw_clear_batch_cache( struct brw_context *brw )
-{
-   struct brw_cached_batch_item *item = brw->cached_batch_items;
-
-   while (item) {
-      struct brw_cached_batch_item *next = item->next;
-      free((void *)item->header);
-      free(item);
-      item = next;
-   }
-
-   brw->cached_batch_items = NULL;
-}
-
-void brw_destroy_batch_cache( struct brw_context *brw )
-{
-   brw_clear_batch_cache(brw);
-}
-
 /**
  * Allocates a block of space in the batchbuffer for indirect state.
  *
diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c
index 08b7ac9079..8bb92fed5d 100644
--- a/src/mesa/drivers/dri/i965/brw_state_upload.c
+++ b/src/mesa/drivers/dri/i965/brw_state_upload.c
@@ -176,7 +176,6 @@ void brw_init_state( struct brw_context *brw )
 void brw_destroy_state( struct brw_context *brw )
 {
    brw_destroy_caches(brw);
-   brw_destroy_batch_cache(brw);
 }
 
 /***********************************************************************
@@ -383,9 +382,6 @@ void brw_validate_state( struct brw_context *brw )
        state->brw == 0)
       return;
 
-   if (brw->state.dirty.brw & BRW_NEW_CONTEXT)
-      brw_clear_batch_cache(brw);
-
    brw->intel.Fallback = GL_FALSE; /* boolean, not bitfield */
 
    /* do prepare stage for all atoms */