i965: don't copy VS attribute work arounds for HSW+
[mesa.git] / src / mesa / drivers / dri / i965 / brw_vs.c
index 7c783f66864abccf82b5988d148f4fe851947503..cda95f81f1b2bf940534d79810d9a7a4dff9016b 100644 (file)
 #include "program/prog_print.h"
 #include "program/prog_parameter.h"
 #include "brw_nir.h"
+#include "brw_program.h"
 
 #include "util/ralloc.h"
 
+GLbitfield64
+brw_vs_outputs_written(struct brw_context *brw, struct brw_vs_prog_key *key,
+                       GLbitfield64 user_varyings)
+{
+   GLbitfield64 outputs_written = user_varyings;
+
+   if (key->copy_edgeflag) {
+      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
+   }
+
+   if (brw->gen < 6) {
+      /* Put dummy slots into the VUE for the SF to put the replaced
+       * point sprite coords in.  We shouldn't need these dummy slots,
+       * which take up precious URB space, but it would mean that the SF
+       * doesn't get nice aligned pairs of input coords into output
+       * coords, which would be a pain to handle.
+       */
+      for (unsigned i = 0; i < 8; i++) {
+         if (key->point_coord_replace & (1 << i))
+            outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
+      }
+
+      /* if back colors are written, allocate slots for front colors too */
+      if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
+         outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
+      if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
+         outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
+   }
+
+   /* In order for legacy clipping to work, we need to populate the clip
+    * distance varying slots whenever clipping is enabled, even if the vertex
+    * shader doesn't write to gl_ClipDistance.
+    */
+   if (key->nr_userclip_plane_consts > 0) {
+      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
+      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
+   }
+
+   return outputs_written;
+}
+
 bool
 brw_codegen_vs_prog(struct brw_context *brw,
                     struct gl_shader_program *prog,
@@ -54,7 +96,6 @@ brw_codegen_vs_prog(struct brw_context *brw,
    struct brw_vs_prog_data prog_data;
    struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
    void *mem_ctx;
-   int i;
    struct brw_shader *vs = NULL;
    bool start_busy = false;
    double start_time = 0;
@@ -79,9 +120,7 @@ brw_codegen_vs_prog(struct brw_context *brw,
     * prog_data associated with the compiled program, and which will be freed
     * by the state cache.
     */
-   int param_count = vp->program.Base.nir->num_uniforms;
-   if (!compiler->scalar_stage[MESA_SHADER_VERTEX])
-      param_count *= 4;
+   int param_count = vp->program.Base.nir->num_uniforms / 4;
 
    if (vs)
       prog_data.base.base.nr_image_params = vs->base.NumImages;
@@ -109,45 +148,23 @@ brw_codegen_vs_prog(struct brw_context *brw,
                                  &prog_data.base.base);
    }
 
-   GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
+   GLbitfield64 outputs_written =
+      brw_vs_outputs_written(brw, key, vp->program.Base.OutputsWritten);
    prog_data.inputs_read = vp->program.Base.InputsRead;
 
    if (key->copy_edgeflag) {
-      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
       prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
    }
 
-   if (brw->gen < 6) {
-      /* Put dummy slots into the VUE for the SF to put the replaced
-       * point sprite coords in.  We shouldn't need these dummy slots,
-       * which take up precious URB space, but it would mean that the SF
-       * doesn't get nice aligned pairs of input coords into output
-       * coords, which would be a pain to handle.
-       */
-      for (i = 0; i < 8; i++) {
-         if (key->point_coord_replace & (1 << i))
-            outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
-      }
-
-      /* if back colors are written, allocate slots for front colors too */
-      if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
-         outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
-      if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
-         outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
-   }
-
-   /* In order for legacy clipping to work, we need to populate the clip
-    * distance varying slots whenever clipping is enabled, even if the vertex
-    * shader doesn't write to gl_ClipDistance.
-    */
-   if (key->nr_userclip_plane_consts > 0) {
-      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
-      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
-   }
+   prog_data.base.cull_distance_mask =
+      ((1 << vp->program.Base.CullDistanceArraySize) - 1) <<
+      vp->program.Base.ClipDistanceArraySize;
 
    brw_compute_vue_map(brw->intelScreen->devinfo,
                        &prog_data.base.vue_map, outputs_written,
-                       prog ? prog->SeparateShader : false);
+                       prog ? prog->SeparateShader ||
+                              prog->_LinkedShaders[MESA_SHADER_TESS_EVAL]
+                            : false);
 
    if (0) {
       _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
@@ -203,11 +220,9 @@ brw_codegen_vs_prog(struct brw_context *brw,
    }
 
    /* Scratch space is used for register spilling */
-   if (prog_data.base.base.total_scratch) {
-      brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
-                        prog_data.base.base.total_scratch *
-                         brw->max_vs_threads);
-   }
+   brw_alloc_stage_scratch(brw, &brw->vs.base,
+                           prog_data.base.base.total_scratch,
+                           brw->max_vs_threads);
 
    brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
                    key, sizeof(struct brw_vs_prog_key),
@@ -219,16 +234,6 @@ brw_codegen_vs_prog(struct brw_context *brw,
    return true;
 }
 
-static bool
-key_debug(struct brw_context *brw, const char *name, int a, int b)
-{
-   if (a != b) {
-      perf_debug("  %s %d->%d\n", name, a, b);
-      return true;
-   }
-   return false;
-}
-
 void
 brw_vs_debug_recompile(struct brw_context *brw,
                        struct gl_shader_program *prog,
@@ -284,7 +289,7 @@ brw_vs_debug_recompile(struct brw_context *brw,
 }
 
 static bool
-brw_vs_state_dirty(struct brw_context *brw)
+brw_vs_state_dirty(const struct brw_context *brw)
 {
    return brw_state_dirty(brw,
                           _NEW_BUFFERS |
@@ -306,7 +311,6 @@ brw_vs_populate_key(struct brw_context *brw,
    struct brw_vertex_program *vp =
       (struct brw_vertex_program *)brw->vertex_program;
    struct gl_program *prog = (struct gl_program *) brw->vertex_program;
-   int i;
 
    memset(key, 0, sizeof(*key));
 
@@ -337,19 +341,17 @@ brw_vs_populate_key(struct brw_context *brw,
 
    /* _NEW_POINT */
    if (brw->gen < 6 && ctx->Point.PointSprite) {
-      for (i = 0; i < 8; i++) {
-        if (ctx->Point.CoordReplace[i])
-            key->point_coord_replace |= (1 << i);
-      }
+      key->point_coord_replace = ctx->Point.CoordReplace & 0xff;
    }
 
    /* _NEW_TEXTURE */
-   brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
-                                      &key->tex);
+   brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 
    /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
-   memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
-          sizeof(brw->vb.attrib_wa_flags));
+   if (brw->gen < 8 && !brw->is_haswell) {
+      memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
+             sizeof(brw->vb.attrib_wa_flags));
+   }
 }
 
 void