svga: update driver for version 10 GPU interface
[mesa.git] / src / gallium / drivers / svga / svga_pipe_sampler.c
index 460a101f8c0f5c8c2cb3d0a3abae99f898857fe5..bb18f5a8247f62cc871e686d1cc6dbe3e81ebef7 100644 (file)
  *
  **********************************************************/
 
-#include "pipe/p_inlines.h"
 #include "pipe/p_defines.h"
+#include "util/u_bitmask.h"
+#include "util/u_format.h"
+#include "util/u_inlines.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
-#include "util/u_pack_color.h"
 #include "tgsi/tgsi_parse.h"
 
 #include "svga_context.h"
-#include "svga_screen_texture.h"
-#include "svga_state.h"
-
-#include "svga_hw_reg.h"
-
+#include "svga_cmd.h"
 #include "svga_debug.h"
+#include "svga_resource_texture.h"
+
 
-static INLINE unsigned
+static inline unsigned
 translate_wrap_mode(unsigned wrap)
 {
    switch (wrap) {
@@ -71,7 +70,7 @@ translate_wrap_mode(unsigned wrap)
    }
 }
 
-static INLINE unsigned translate_img_filter( unsigned filter )
+static inline unsigned translate_img_filter( unsigned filter )
 {
    switch (filter) {
    case PIPE_TEX_FILTER_NEAREST: return SVGA3D_TEX_FILTER_NEAREST;
@@ -82,7 +81,7 @@ static INLINE unsigned translate_img_filter( unsigned filter )
    }
 }
 
-static INLINE unsigned translate_mip_filter( unsigned filter )
+static inline unsigned translate_mip_filter( unsigned filter )
 {
    switch (filter) {
    case PIPE_TEX_MIPFILTER_NONE:    return SVGA3D_TEX_FILTER_NONE;
@@ -94,19 +93,141 @@ static INLINE unsigned translate_mip_filter( unsigned filter )
    }
 }
 
+
+static uint8
+translate_comparison_func(unsigned func)
+{
+   switch (func) {
+   case PIPE_FUNC_NEVER:
+      return SVGA3D_COMPARISON_NEVER;
+   case PIPE_FUNC_LESS:
+      return SVGA3D_COMPARISON_LESS;
+   case PIPE_FUNC_EQUAL:
+      return SVGA3D_COMPARISON_EQUAL;
+   case PIPE_FUNC_LEQUAL:
+      return SVGA3D_COMPARISON_LESS_EQUAL;
+   case PIPE_FUNC_GREATER:
+      return SVGA3D_COMPARISON_GREATER;
+   case PIPE_FUNC_NOTEQUAL:
+      return SVGA3D_COMPARISON_NOT_EQUAL;
+   case PIPE_FUNC_GEQUAL:
+      return SVGA3D_COMPARISON_GREATER_EQUAL;
+   case PIPE_FUNC_ALWAYS:
+      return SVGA3D_COMPARISON_ALWAYS;
+   default:
+      assert(!"Invalid comparison function");
+      return SVGA3D_COMPARISON_ALWAYS;
+   }
+}
+
+
+/**
+ * Translate filtering state to vgpu10 format.
+ */
+static SVGA3dFilter
+translate_filter_mode(unsigned img_filter,
+                      unsigned min_filter,
+                      unsigned mag_filter,
+                      boolean anisotropic,
+                      boolean compare)
+{
+   SVGA3dFilter mode = 0;
+
+   if (img_filter == PIPE_TEX_FILTER_LINEAR)
+      mode |= SVGA3D_FILTER_MIP_LINEAR;
+   if (min_filter == PIPE_TEX_FILTER_LINEAR)
+      mode |= SVGA3D_FILTER_MIN_LINEAR;
+   if (mag_filter == PIPE_TEX_FILTER_LINEAR)
+      mode |= SVGA3D_FILTER_MAG_LINEAR;
+   if (anisotropic)
+      mode |= SVGA3D_FILTER_ANISOTROPIC;
+   if (compare)
+      mode |= SVGA3D_FILTER_COMPARE;
+
+   return mode;
+}
+
+
+/**
+ * Define a vgpu10 sampler state.
+ */
+static void
+define_sampler_state_object(struct svga_context *svga,
+                            struct svga_sampler_state *ss,
+                            const struct pipe_sampler_state *ps)
+{
+   uint8_t max_aniso = (uint8_t) 255; /* XXX fix me */
+   boolean anisotropic;
+   uint8 compare_func;
+   SVGA3dFilter filter;
+   SVGA3dRGBAFloat bcolor;
+   unsigned try;
+   float min_lod, max_lod;
+
+   assert(svga_have_vgpu10(svga));
+
+   anisotropic = ss->aniso_level > 1.0f;
+
+   filter = translate_filter_mode(ps->min_mip_filter,
+                                  ps->min_img_filter,
+                                  ps->mag_img_filter,
+                                  anisotropic,
+                                  ss->compare_mode);
+
+   compare_func = translate_comparison_func(ss->compare_func);
+
+   COPY_4V(bcolor.value, ps->border_color.f);
+
+   ss->id = util_bitmask_add(svga->sampler_object_id_bm);
+
+   assert(ps->min_lod <= ps->max_lod);
+
+   if (ps->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
+      /* just use the base level image */
+      min_lod = max_lod = 0.0f;
+   }
+   else {
+      min_lod = ps->min_lod;
+      max_lod = ps->max_lod;
+   }
+
+   /* Loop in case command buffer is full and we need to flush and retry */
+   for (try = 0; try < 2; try++) {
+      enum pipe_error ret =
+         SVGA3D_vgpu10_DefineSamplerState(svga->swc,
+                                          ss->id,
+                                          filter,
+                                          ss->addressu,
+                                          ss->addressv,
+                                          ss->addressw,
+                                          ss->lod_bias, /* float */
+                                          max_aniso,
+                                          compare_func,
+                                          bcolor,
+                                          min_lod,       /* float */
+                                          max_lod);      /* float */
+      if (ret == PIPE_OK)
+         return;
+      svga_context_flush(svga, NULL);
+   }
+}
+
+
 static void *
 svga_create_sampler_state(struct pipe_context *pipe,
                           const struct pipe_sampler_state *sampler)
 {
    struct svga_context *svga = svga_context(pipe);
    struct svga_sampler_state *cso = CALLOC_STRUCT( svga_sampler_state );
-   union util_color uc;
    
+   if (!cso)
+      return NULL;
+
    cso->mipfilter = translate_mip_filter(sampler->min_mip_filter);
    cso->magfilter = translate_img_filter( sampler->mag_img_filter );
    cso->minfilter = translate_img_filter( sampler->min_img_filter );
-   cso->aniso_level = MAX2( (unsigned) sampler->max_anisotropy, 1 );
-   if(cso->aniso_level != 1)
+   cso->aniso_level = MAX2( sampler->max_anisotropy, 1 );
+   if(sampler->max_anisotropy)
       cso->magfilter = cso->minfilter = SVGA3D_TEX_FILTER_ANISOTROPIC;
    cso->lod_bias = sampler->lod_bias;
    cso->addressu = translate_wrap_mode(sampler->wrap_s);
@@ -117,22 +238,20 @@ svga_create_sampler_state(struct pipe_context *pipe,
    cso->compare_func = sampler->compare_func;
 
    {
-      ubyte r = float_to_ubyte(sampler->border_color[0]);
-      ubyte g = float_to_ubyte(sampler->border_color[1]);
-      ubyte b = float_to_ubyte(sampler->border_color[2]);
-      ubyte a = float_to_ubyte(sampler->border_color[3]);
+      uint32 r = float_to_ubyte(sampler->border_color.f[0]);
+      uint32 g = float_to_ubyte(sampler->border_color.f[1]);
+      uint32 b = float_to_ubyte(sampler->border_color.f[2]);
+      uint32 a = float_to_ubyte(sampler->border_color.f[3]);
 
-      util_pack_color_ub( r, g, b, a,
-                          PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
-      cso->bordercolor = uc.ui;
+      cso->bordercolor = (a << 24) | (r << 16) | (g << 8) | b;
    }
 
    /* No SVGA3D support for:
     *    - min/max LOD clamping
     */
    cso->min_lod = 0;
-   cso->view_min_lod = MAX2(sampler->min_lod, 0);
-   cso->view_max_lod = MAX2(sampler->max_lod, 0);
+   cso->view_min_lod = MAX2((int) (sampler->min_lod + 0.5), 0);
+   cso->view_max_lod = MAX2((int) (sampler->max_lod + 0.5), 0);
 
    /* Use min_mipmap */
    if (svga->debug.use_min_mipmap) {
@@ -144,6 +263,10 @@ svga_create_sampler_state(struct pipe_context *pipe,
       }
    }
 
+   if (svga_have_vgpu10(svga)) {
+      define_sampler_state_object(svga, cso, sampler);
+   }
+
    SVGA_DBG(DEBUG_VIEWS, "min %u, view(min %u, max %u) lod, mipfilter %s\n",
             cso->min_lod, cso->view_min_lod, cso->view_max_lod,
             cso->mipfilter == SVGA3D_TEX_FILTER_NONE ? "SVGA3D_TEX_FILTER_NONE" : "SOMETHING");
@@ -151,75 +274,169 @@ svga_create_sampler_state(struct pipe_context *pipe,
    return cso;
 }
 
-static void svga_bind_sampler_states(struct pipe_context *pipe,
-                                     unsigned num, void **sampler)
+static void
+svga_bind_sampler_states(struct pipe_context *pipe,
+                         unsigned shader,
+                         unsigned start,
+                         unsigned num,
+                         void **samplers)
 {
    struct svga_context *svga = svga_context(pipe);
    unsigned i;
 
-   assert(num <= PIPE_MAX_SAMPLERS);
+   assert(shader < PIPE_SHADER_TYPES);
+   assert(start + num <= PIPE_MAX_SAMPLERS);
 
-   /* Check for no-op */
-   if (num == svga->curr.num_samplers &&
-       !memcmp(svga->curr.sampler, sampler, num * sizeof(void *))) {
-      debug_printf("sampler noop\n");
+   /* Pre-VGPU10 only supports FS textures */
+   if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
       return;
-   }
 
    for (i = 0; i < num; i++)
-      svga->curr.sampler[i] = sampler[i];
+      svga->curr.sampler[shader][start + i] = samplers[i];
 
-   for (i = num; i < svga->curr.num_samplers; i++)
-      svga->curr.sampler[i] = NULL;
+   /* find highest non-null sampler[] entry */
+   {
+      unsigned j = MAX2(svga->curr.num_samplers[shader], start + num);
+      while (j > 0 && svga->curr.sampler[shader][j - 1] == NULL)
+         j--;
+      svga->curr.num_samplers[shader] = j;
+   }
 
-   svga->curr.num_samplers = num;
    svga->dirty |= SVGA_NEW_SAMPLER;
 }
 
+
 static void svga_delete_sampler_state(struct pipe_context *pipe,
                                       void *sampler)
 {
+   struct svga_sampler_state *ss = (struct svga_sampler_state *) sampler;
+   struct svga_context *svga = svga_context(pipe);
+
+   if (svga_have_vgpu10(svga)) {
+      enum pipe_error ret;
+
+      svga_hwtnl_flush_retry(svga);
+
+      ret = SVGA3D_vgpu10_DestroySamplerState(svga->swc, ss->id);
+      if (ret != PIPE_OK) {
+         svga_context_flush(svga, NULL);
+         ret = SVGA3D_vgpu10_DestroySamplerState(svga->swc, ss->id);
+      }
+      util_bitmask_clear(svga->sampler_object_id_bm, ss->id);
+   }
+
    FREE(sampler);
 }
 
 
-static void svga_set_sampler_textures(struct pipe_context *pipe,
-                                      unsigned num,
-                                      struct pipe_texture **texture)
+static struct pipe_sampler_view *
+svga_create_sampler_view(struct pipe_context *pipe,
+                         struct pipe_resource *texture,
+                         const struct pipe_sampler_view *templ)
+{
+   struct svga_pipe_sampler_view *sv = CALLOC_STRUCT(svga_pipe_sampler_view);
+
+   if (!sv) {
+      return NULL;
+   }
+
+   sv->base = *templ;
+   sv->base.reference.count = 1;
+   sv->base.texture = NULL;
+   pipe_resource_reference(&sv->base.texture, texture);
+
+   sv->base.context = pipe;
+   sv->id = SVGA3D_INVALID_ID;
+
+   return &sv->base;
+}
+
+
+static void
+svga_sampler_view_destroy(struct pipe_context *pipe,
+                          struct pipe_sampler_view *view)
+{
+   struct svga_context *svga = svga_context(pipe);
+   struct svga_pipe_sampler_view *sv = svga_pipe_sampler_view(view);
+
+   if (svga_have_vgpu10(svga) && sv->id != SVGA3D_INVALID_ID) {
+      if (view->context != pipe) {
+         /* The SVGA3D device will generate an error (and on Linux, cause
+          * us to abort) if we try to destroy a shader resource view from
+          * a context other than the one it was created with.  Skip the
+          * SVGA3D_vgpu10_DestroyShaderResourceView() and leak the sampler
+          * view for now.  This should only sometimes happen when a shared
+          * texture is deleted.
+          */
+         _debug_printf("context mismatch in %s\n", __func__);
+      }
+      else {
+         enum pipe_error ret;
+
+         svga_hwtnl_flush_retry(svga); /* XXX is this needed? */
+
+         ret = SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc, sv->id);
+         if (ret != PIPE_OK) {
+            svga_context_flush(svga, NULL);
+            ret = SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc, sv->id);
+         }
+         util_bitmask_clear(svga->sampler_view_id_bm, sv->id);
+      }
+   }
+
+   pipe_resource_reference(&sv->base.texture, NULL);
+
+   FREE(sv);
+}
+
+static void
+svga_set_sampler_views(struct pipe_context *pipe,
+                       unsigned shader,
+                       unsigned start,
+                       unsigned num,
+                       struct pipe_sampler_view **views)
 {
    struct svga_context *svga = svga_context(pipe);
    unsigned flag_1d = 0;
    unsigned flag_srgb = 0;
    uint i;
 
-   assert(num <= PIPE_MAX_SAMPLERS);
+   assert(shader < PIPE_SHADER_TYPES);
+   assert(start + num <= Elements(svga->curr.sampler_views[shader]));
 
-   /* Check for no-op */
-   if (num == svga->curr.num_textures &&
-       !memcmp(svga->curr.texture, texture, num * sizeof(struct pipe_texture *))) {
-      if (0) debug_printf("texture noop\n");
+   /* Pre-VGPU10 only supports FS textures */
+   if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
       return;
-   }
 
    for (i = 0; i < num; i++) {
-      pipe_texture_reference(&svga->curr.texture[i],
-                             texture[i]);
+      if (svga->curr.sampler_views[shader][start + i] != views[i]) {
+         /* Note: we're using pipe_sampler_view_release() here to work around
+          * a possible crash when the old view belongs to another context that
+          * was already destroyed.
+          */
+         pipe_sampler_view_release(pipe, &svga->curr.sampler_views[shader][start + i]);
+         pipe_sampler_view_reference(&svga->curr.sampler_views[shader][start + i],
+                                     views[i]);
+      }
 
-      if (!texture[i])
+      if (!views[i])
          continue;
 
-      if (texture[i]->format == PIPE_FORMAT_A8R8G8B8_SRGB)
-         flag_srgb |= 1 << i;
+      if (util_format_is_srgb(views[i]->format))
+         flag_srgb |= 1 << (start + i);
 
-      if (texture[i]->target == PIPE_TEXTURE_1D)
-         flag_1d |= 1 << i;
+      if (views[i]->texture->target == PIPE_TEXTURE_1D)
+         flag_1d |= 1 << (start + i);
    }
 
-   for (i = num; i < svga->curr.num_textures; i++)
-      pipe_texture_reference(&svga->curr.texture[i],
-                             NULL);
+   /* find highest non-null sampler_views[] entry */
+   {
+      unsigned j = MAX2(svga->curr.num_sampler_views[shader], start + num);
+      while (j > 0 && svga->curr.sampler_views[shader][j - 1] == NULL)
+         j--;
+      svga->curr.num_sampler_views[shader] = j;
+   }
 
-   svga->curr.num_textures = num;
    svga->dirty |= SVGA_NEW_TEXTURE_BINDING;
 
    if (flag_srgb != svga->curr.tex_flags.flag_srgb ||
@@ -232,13 +449,14 @@ static void svga_set_sampler_textures(struct pipe_context *pipe,
 }
 
 
-
 void svga_init_sampler_functions( struct svga_context *svga )
 {
    svga->pipe.create_sampler_state = svga_create_sampler_state;
-   svga->pipe.bind_fragment_sampler_states = svga_bind_sampler_states;
+   svga->pipe.bind_sampler_states = svga_bind_sampler_states;
    svga->pipe.delete_sampler_state = svga_delete_sampler_state;
-   svga->pipe.set_fragment_sampler_textures = svga_set_sampler_textures;
+   svga->pipe.set_sampler_views = svga_set_sampler_views;
+   svga->pipe.create_sampler_view = svga_create_sampler_view;
+   svga->pipe.sampler_view_destroy = svga_sampler_view_destroy;
 }