svga: update constant buffer code for GBS

Reviewed-by: Thomas Hellstrom <[email protected]>
Cc: "10.1" <[email protected]>

5 files changed, 175 insertions, 69 deletions

diff --git a/src/gallium/drivers/svga/svga_context.c b/src/gallium/drivers/svga/svga_context.c
index c32051a41db..de769cac95f 100644
--- a/src/gallium/drivers/svga/svga_context.c
+++ b/src/gallium/drivers/svga/svga_context.c
@@ -53,6 +53,7 @@ DEBUG_GET_ONCE_BOOL_OPTION(force_hw_line_stipple, "SVGA_FORCE_HW_LINE_STIPPLE",
 static void svga_destroy( struct pipe_context *pipe )
 {
    struct svga_context *svga = svga_context( pipe );
+   struct svga_winsys_screen *sws = svga_screen(pipe->screen)->sws;
    unsigned shader;
 
    util_blitter_destroy(svga->blitter);
@@ -70,8 +71,10 @@ static void svga_destroy( struct pipe_context *pipe )
 
    util_bitmask_destroy( svga->shader_id_bm );
 
-   for(shader = 0; shader < PIPE_SHADER_TYPES; ++shader)
-      pipe_resource_reference( &svga->curr.cb[shader], NULL );
+   for (shader = 0; shader < PIPE_SHADER_TYPES; ++shader) {
+      pipe_resource_reference( &svga->curr.cbufs[shader].buffer, NULL );
+      sws->surface_reference(sws, &svga->state.hw_draw.hw_cb[shader], NULL);
+   }
 
    FREE( svga );
 }
@@ -147,6 +150,7 @@ struct pipe_context *svga_context_create( struct pipe_screen *screen,
    memset(&svga->state.hw_draw, 0xcd, sizeof(svga->state.hw_draw));
    memset(&svga->state.hw_draw.views, 0x0, sizeof(svga->state.hw_draw.views));
    svga->state.hw_draw.num_views = 0;
+   memset(&svga->state.hw_draw.hw_cb, 0x0, sizeof(svga->state.hw_draw.hw_cb));
 
    svga->dirty = ~0;
 
diff --git a/src/gallium/drivers/svga/svga_context.h b/src/gallium/drivers/svga/svga_context.h
index 5b04bb482a1..71a8eea3fe1 100644
--- a/src/gallium/drivers/svga/svga_context.h
+++ b/src/gallium/drivers/svga/svga_context.h
@@ -219,7 +219,7 @@ struct svga_state
 
    struct pipe_vertex_buffer vb[PIPE_MAX_ATTRIBS];
    struct pipe_index_buffer ib;
-   struct pipe_resource *cb[PIPE_SHADER_TYPES];
+   struct pipe_constant_buffer cbufs[PIPE_SHADER_TYPES];
 
    struct pipe_framebuffer_state framebuffer;
    float depthscale;
@@ -287,6 +287,11 @@ struct svga_hw_draw_state
    unsigned ts[SVGA3D_PIXEL_SAMPLERREG_MAX][SVGA3D_TS_MAX];
    float cb[PIPE_SHADER_TYPES][SVGA3D_CONSTREG_MAX][4];
 
+   /**
+    * For guest backed shader constants only.
+    */
+   struct svga_winsys_surface *hw_cb[PIPE_SHADER_TYPES];
+
    struct svga_shader_variant *fs;
    struct svga_shader_variant *vs;
    struct svga_hw_view_state views[PIPE_MAX_SAMPLERS];
diff --git a/src/gallium/drivers/svga/svga_pipe_constants.c b/src/gallium/drivers/svga/svga_pipe_constants.c
index cfa823b345d..c32b66d416b 100644
--- a/src/gallium/drivers/svga/svga_pipe_constants.c
+++ b/src/gallium/drivers/svga/svga_pipe_constants.c
@@ -61,8 +61,10 @@ static void svga_set_constant_buffer(struct pipe_context *pipe,
    assert(shader < PIPE_SHADER_TYPES);
    assert(index == 0);
 
-   pipe_resource_reference( &svga->curr.cb[shader],
-                          buf );
+   pipe_resource_reference(&svga->curr.cbufs[shader].buffer, buf);
+   svga->curr.cbufs[shader].buffer_size = cb ? cb->buffer_size : 0;
+   svga->curr.cbufs[shader].buffer_offset = cb ? cb->buffer_offset : 0;
+   svga->curr.cbufs[shader].user_buffer = NULL; /* not used */
 
    if (shader == PIPE_SHADER_FRAGMENT)
       svga->dirty |= SVGA_NEW_FS_CONST_BUFFER;
diff --git a/src/gallium/drivers/svga/svga_state_constants.c b/src/gallium/drivers/svga/svga_state_constants.c
index d4682981279..1e1fbb099c6 100644
--- a/src/gallium/drivers/svga/svga_state_constants.c
+++ b/src/gallium/drivers/svga/svga_state_constants.c
@@ -24,6 +24,7 @@
  **********************************************************/
 
 #include "util/u_inlines.h"
+#include "util/u_memory.h"
 #include "pipe/p_defines.h"
 
 #include "svga_screen.h"
@@ -32,6 +33,7 @@
 #include "svga_cmd.h"
 #include "svga_tgsi.h"
 #include "svga_debug.h"
+#include "svga_resource_buffer.h"
 
 #include "svga_hw_reg.h"
 
@@ -41,7 +43,14 @@
  */
 #define MAX_CONST_REG_COUNT 256  /**< number of float[4] constants */
 
+/**
+ * Extra space for svga-specific VS/PS constants (such as texcoord
+ * scale factors, vertex transformation scale/translation).
+ */
+#define MAX_EXTRA_CONSTS 32
 
+/** Guest-backed surface constant buffers must be this size */
+#define GB_CONSTBUF_SIZE (SVGA3D_CONSTREG_MAX)
 
 /**
  * Convert from PIPE_SHADER_* to SVGA3D_SHADERTYPE_*
@@ -62,6 +71,81 @@ svga_shader_type(unsigned shader)
 
 
 /**
+ * Emit any extra fragment shader constants into the buffer pointed
+ * to by 'dest'.
+ * In particular, these would be the scaling factors needed for handling
+ * unnormalized texture coordinates for texture rectangles.
+ * \return number of float[4] constants put into the dest buffer
+ */
+static unsigned
+svga_get_extra_fs_constants(struct svga_context *svga, float *dest)
+{
+   const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
+   const struct svga_fs_compile_key *key = &variant->key.fkey;
+   unsigned count = 0;
+
+   /* SVGA_NEW_VS_VARIANT
+    */
+   if (key->num_unnormalized_coords) {
+      unsigned i;
+
+      for (i = 0; i < key->num_textures; i++) {
+         if (key->tex[i].unnormalized) {
+            struct pipe_resource *tex = svga->curr.sampler_views[i]->texture;
+
+            /* debug/sanity check */
+            assert(key->tex[i].width_height_idx == count);
+
+            *dest++ = 1.0 / (float)tex->width0;
+            *dest++ = 1.0 / (float)tex->height0;
+            *dest++ = 1.0;
+            *dest++ = 1.0;
+
+            count++;
+         }
+      }
+   }
+
+   assert(count <= MAX_EXTRA_CONSTS);
+
+   return count;
+}
+
+
+/**
+ * Emit any extra vertex shader constants into the buffer pointed
+ * to by 'dest'.
+ * In particular, these would be the scale and bias factors computed
+ * from the framebuffer size which are used to copy with differences in
+ * GL vs D3D coordinate spaces.  See svga_tgsi_insn.c for more info.
+ * \return number of float[4] constants put into the dest buffer
+ */
+static unsigned
+svga_get_extra_vs_constants(struct svga_context *svga, float *dest)
+{
+   const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
+   const struct svga_vs_compile_key *key = &variant->key.vkey;
+   unsigned count = 0;
+
+   /* SVGA_NEW_VS_VARIANT
+    */
+   if (key->need_prescale) {
+      memcpy(dest, svga->state.hw_clear.prescale.scale, 4 * sizeof(float));
+      dest += 4;
+
+      memcpy(dest, svga->state.hw_clear.prescale.translate, 4 * sizeof(float));
+      dest += 4;
+
+      count = 2;
+   }
+
+   assert(count <= MAX_EXTRA_CONSTS);
+
+   return count;
+}
+
+
+/**
  * Check and emit one shader constant register.
  * \param shader  PIPE_SHADER_FRAGMENT or PIPE_SHADER_VERTEX
  * \param i  which float[4] constant to change
@@ -120,8 +204,8 @@ emit_const_range(struct svga_context *svga,
 
 #ifdef DEBUG
    if (offset + count > SVGA3D_CONSTREG_MAX) {
-      debug_printf("svga: too many constants (offset + count = %u)\n",
-                   offset + count);
+      debug_printf("svga: too many constants (offset %u + count %u = %u (max = %u))\n",
+                   offset, count, offset + count, SVGA3D_CONSTREG_MAX);
    }
 #endif
 
@@ -185,11 +269,21 @@ emit_const_range(struct svga_context *svga,
 
          /* Send them all together.
           */
-         ret = SVGA3D_SetShaderConsts(svga->swc,
-                                      offset + i, j - i,
-                                      svga_shader_type(shader),
-                                      SVGA3D_CONST_TYPE_FLOAT,
-                                      values + i);
+         if (svga_have_gb_objects(svga)) {
+            ret = SVGA3D_SetGBShaderConstsInline(svga->swc,
+                                                 offset + i, /* start */
+                                                 j - i,  /* count */
+                                                 svga_shader_type(shader),
+                                                 SVGA3D_CONST_TYPE_FLOAT,
+                                                 values + i);
+         }
+         else {
+            ret = SVGA3D_SetShaderConsts(svga->swc,
+                                         offset + i, j - i,
+                                         svga_shader_type(shader),
+                                         SVGA3D_CONST_TYPE_FLOAT,
+                                         values + i);
+         }
          if (ret != PIPE_OK) {
             return ret;
          }
@@ -227,13 +321,11 @@ emit_consts(struct svga_context *svga, unsigned shader)
 
    assert(shader < PIPE_SHADER_TYPES);
 
-   if (svga->curr.cb[shader] == NULL)
+   if (svga->curr.cbufs[shader].buffer == NULL)
       goto done;
 
-   count = svga->curr.cb[shader]->width0 / (4 * sizeof(float));
-
    data = (const float (*)[4])pipe_buffer_map(&svga->pipe,
-                                              svga->curr.cb[shader],
+                                              svga->curr.cbufs[shader].buffer,
                                               PIPE_TRANSFER_READ,
 					      &transfer);
    if (data == NULL) {
@@ -241,6 +333,14 @@ emit_consts(struct svga_context *svga, unsigned shader)
       goto done;
    }
 
+   /* sanity check */
+   assert(svga->curr.cbufs[shader].buffer->width0 >=
+          svga->curr.cbufs[shader].buffer_size);
+
+   /* Use/apply the constant buffer size and offsets here */
+   count = svga->curr.cbufs[shader].buffer_size / (4 * sizeof(float));
+   data += svga->curr.cbufs[shader].buffer_offset / (4 * sizeof(float));
+
    if (ss->hw_version >= SVGA3D_HWVERSION_WS8_B1) {
       ret = emit_const_range( svga, shader, offset, count, data );
       if (ret != PIPE_OK) {
@@ -266,47 +366,42 @@ done:
 static enum pipe_error
 emit_fs_consts(struct svga_context *svga, unsigned dirty)
 {
+   struct svga_screen *ss = svga_screen(svga->pipe.screen);
    const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
    enum pipe_error ret = PIPE_OK;
 
+   /* SVGA_NEW_FS_VARIANT
+    */
+   if (variant == NULL)
+      return PIPE_OK;
+
+   /* SVGA_NEW_FS_CONST_BUFFER
+    */
    ret = emit_consts( svga, PIPE_SHADER_FRAGMENT );
    if (ret != PIPE_OK)
       return ret;
 
-   /* The internally generated fragment shader for xor blending
-    * doesn't have a 'variant' struct.  It should be fixed to avoid
-    * this special case, but work around it with a NULL check:
-    */
-   if (variant) {
-      const struct svga_fs_compile_key *key = &variant->key.fkey;
-      if (key->num_unnormalized_coords) {
-         const unsigned offset =
-            variant->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
-         unsigned i;
-
-         for (i = 0; i < key->num_textures; i++) {
-            if (key->tex[i].unnormalized) {
-               struct pipe_resource *tex = svga->curr.sampler_views[i]->texture;
-               float data[4];
-
-               data[0] = 1.0f / (float) tex->width0;
-               data[1] = 1.0f / (float) tex->height0;
-               data[2] = 1.0f;
-               data[3] = 1.0f;
-
-               ret = emit_const(svga,
-                                PIPE_SHADER_FRAGMENT,
-                                key->tex[i].width_height_idx + offset,
-                                data);
-               if (ret != PIPE_OK) {
-                  return ret;
-               }
-            }
+   /* emit extra shader constants */
+   {
+      unsigned offset = variant->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
+      float extras[MAX_EXTRA_CONSTS][4];
+      unsigned count, i;
+
+      count = svga_get_extra_fs_constants(svga, (float *) extras);
+
+      if (ss->hw_version >= SVGA3D_HWVERSION_WS8_B1) {
+         ret = emit_const_range(svga, PIPE_SHADER_FRAGMENT, offset, count,
+                                (const float (*) [4])extras);
+      } else {
+         for (i = 0; i < count; i++) {
+            ret = emit_const(svga, PIPE_SHADER_FRAGMENT, offset + i, extras[i]);
+            if (ret != PIPE_OK)
+               return ret;
          }
       }
    }
 
-   return PIPE_OK;
+   return ret;
 }
 
 
@@ -324,43 +419,43 @@ struct svga_tracked_state svga_hw_fs_constants =
 static enum pipe_error
 emit_vs_consts(struct svga_context *svga, unsigned dirty)
 {
+   struct svga_screen *ss = svga_screen(svga->pipe.screen);
    const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
-   const struct svga_vs_compile_key *key;
    enum pipe_error ret = PIPE_OK;
-   unsigned offset;
 
    /* SVGA_NEW_VS_VARIANT
     */
    if (variant == NULL)
       return PIPE_OK;
 
-   key = &variant->key.vkey;
-
    /* SVGA_NEW_VS_CONST_BUFFER
     */
    ret = emit_consts( svga, PIPE_SHADER_VERTEX );
    if (ret != PIPE_OK)
       return ret;
 
-   /* offset = number of constants in the VS const buffer */
-   offset = variant->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
+   /* emit extra shader constants */
+   {
+      unsigned offset = variant->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
+      float extras[MAX_EXTRA_CONSTS][4];
+      unsigned count, i;
 
-   /* SVGA_NEW_VS_PRESCALE
-    * Put the viewport pre-scale/translate values into the const buffer.
-    */
-   if (key->need_prescale) {
-      ret = emit_const( svga, PIPE_SHADER_VERTEX, offset++,
-                        svga->state.hw_clear.prescale.scale );
-      if (ret != PIPE_OK)
-         return ret;
+      count = svga_get_extra_vs_constants(svga, (float *) extras);
+      assert(count <= Elements(extras));
 
-      ret = emit_const( svga, PIPE_SHADER_VERTEX, offset++,
-                        svga->state.hw_clear.prescale.translate );
-      if (ret != PIPE_OK)
-         return ret;
+      if (ss->hw_version >= SVGA3D_HWVERSION_WS8_B1) {
+         ret = emit_const_range(svga, PIPE_SHADER_VERTEX, offset, count,
+                                (const float (*) [4]) extras);
+      } else {
+         for (i = 0; i < count; i++) {
+            ret = emit_const(svga, PIPE_SHADER_VERTEX, offset + i, extras[i]);
+            if (ret != PIPE_OK)
+               return ret;
+         }
+      }
    }
 
-   return PIPE_OK;
+   return ret;
 }
 
 
diff --git a/src/gallium/drivers/svga/svga_swtnl_draw.c b/src/gallium/drivers/svga/svga_swtnl_draw.c
index de38d6f59f3..832249523cc 100644
--- a/src/gallium/drivers/svga/svga_swtnl_draw.c
+++ b/src/gallium/drivers/svga/svga_swtnl_draw.c
@@ -88,16 +88,16 @@ svga_swtnl_draw_vbo(struct svga_context *svga,
                        svga->curr.ib.index_size, ~0);
    }
 
-   if (svga->curr.cb[PIPE_SHADER_VERTEX]) {
+   if (svga->curr.cbufs[PIPE_SHADER_VERTEX].buffer) {
       map = pipe_buffer_map(&svga->pipe,
-                            svga->curr.cb[PIPE_SHADER_VERTEX],
+                            svga->curr.cbufs[PIPE_SHADER_VERTEX].buffer,
                             PIPE_TRANSFER_READ,
 			    &cb_transfer);
       assert(map);
       draw_set_mapped_constant_buffer(
          draw, PIPE_SHADER_VERTEX, 0,
          map,
-         svga->curr.cb[PIPE_SHADER_VERTEX]->width0);
+         svga->curr.cbufs[PIPE_SHADER_VERTEX].buffer->width0);
    }
 
    draw_vbo(draw, info);
@@ -122,7 +122,7 @@ svga_swtnl_draw_vbo(struct svga_context *svga,
       draw_set_indexes(draw, NULL, 0, 0);
    }
 
-   if (svga->curr.cb[PIPE_SHADER_VERTEX]) {
+   if (svga->curr.cbufs[PIPE_SHADER_VERTEX].buffer) {
       pipe_buffer_unmap(&svga->pipe, cb_transfer);
    }