i965/nir/vec4: Implement conditional statements (nir_cf_node_if)
[mesa.git] / src / mesa / drivers / dri / i965 / brw_wm.c
index dfd008ee416d0c95f978fb01cab1c58db23e7482..b590b177ece0a60cce123eb95b37374ba4a30c72 100644 (file)
 #include "main/formats.h"
 #include "main/fbobject.h"
 #include "main/samplerobj.h"
+#include "main/framebuffer.h"
 #include "program/prog_parameter.h"
 #include "program/program.h"
 #include "intel_mipmap_tree.h"
 
-#include "glsl/ralloc.h"
+#include "util/ralloc.h"
 
 /**
  * Return a bitfield where bit n is set if barycentric interpolation mode n
@@ -116,6 +117,25 @@ brw_compute_barycentric_interp_modes(struct brw_context *brw,
    return barycentric_interp_modes;
 }
 
+static uint8_t
+computed_depth_mode(struct gl_fragment_program *fp)
+{
+   if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
+      switch (fp->FragDepthLayout) {
+      case FRAG_DEPTH_LAYOUT_NONE:
+      case FRAG_DEPTH_LAYOUT_ANY:
+         return BRW_PSCDEPTH_ON;
+      case FRAG_DEPTH_LAYOUT_GREATER:
+         return BRW_PSCDEPTH_ON_GE;
+      case FRAG_DEPTH_LAYOUT_LESS:
+         return BRW_PSCDEPTH_ON_LE;
+      case FRAG_DEPTH_LAYOUT_UNCHANGED:
+         return BRW_PSCDEPTH_OFF;
+      }
+   }
+   return BRW_PSCDEPTH_OFF;
+}
+
 bool
 brw_wm_prog_data_compare(const void *in_a, const void *in_b)
 {
@@ -140,14 +160,15 @@ brw_wm_prog_data_compare(const void *in_a, const void *in_b)
  * Depending on the instructions used (i.e. flow control instructions)
  * we'll use one of two code generators.
  */
-bool do_wm_prog(struct brw_context *brw,
-               struct gl_shader_program *prog,
-               struct brw_fragment_program *fp,
-               struct brw_wm_prog_key *key)
+bool
+brw_codegen_wm_prog(struct brw_context *brw,
+                    struct gl_shader_program *prog,
+                    struct brw_fragment_program *fp,
+                    struct brw_wm_prog_key *key)
 {
    struct gl_context *ctx = &brw->ctx;
    void *mem_ctx = ralloc_context(NULL);
-   struct brw_wm_compile *c;
+   struct brw_wm_prog_data prog_data;
    const GLuint *program;
    struct gl_shader *fs = NULL;
    GLuint program_size;
@@ -155,7 +176,18 @@ bool do_wm_prog(struct brw_context *brw,
    if (prog)
       fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 
-   c = rzalloc(mem_ctx, struct brw_wm_compile);
+   memset(&prog_data, 0, sizeof(prog_data));
+   /* key->alpha_test_func means simulating alpha testing via discards,
+    * so the shader definitely kills pixels.
+    */
+   prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
+   prog_data.uses_omask =
+      fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
+   prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
+
+   /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
+   if (!prog)
+      prog_data.base.use_alt_mode = true;
 
    /* Allocate the references to the uniforms that will end up in the
     * prog_data associated with the compiled program, and which will be freed
@@ -169,35 +201,36 @@ bool do_wm_prog(struct brw_context *brw,
    }
    /* The backend also sometimes adds params for texture size. */
    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
-   c->prog_data.base.param = rzalloc_array(NULL, const float *, param_count);
-   c->prog_data.base.pull_param =
-      rzalloc_array(NULL, const float *, param_count);
-   c->prog_data.base.nr_params = param_count;
+   prog_data.base.param =
+      rzalloc_array(NULL, const gl_constant_value *, param_count);
+   prog_data.base.pull_param =
+      rzalloc_array(NULL, const gl_constant_value *, param_count);
+   prog_data.base.nr_params = param_count;
 
-   c->prog_data.barycentric_interp_modes =
+   prog_data.barycentric_interp_modes =
       brw_compute_barycentric_interp_modes(brw, key->flat_shade,
                                            key->persample_shading,
                                            &fp->program);
 
-   program = brw_wm_fs_emit(brw, mem_ctx, key, &c->prog_data,
+   program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
                             &fp->program, prog, &program_size);
    if (program == NULL) {
       ralloc_free(mem_ctx);
       return false;
    }
 
-   if (c->prog_data.total_scratch) {
+   if (prog_data.base.total_scratch) {
       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
-                        c->prog_data.total_scratch * brw->max_wm_threads);
+                        prog_data.base.total_scratch * brw->max_wm_threads);
    }
 
    if (unlikely(INTEL_DEBUG & DEBUG_WM))
       fprintf(stderr, "\n");
 
-   brw_upload_cache(&brw->cache, BRW_WM_PROG,
+   brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
                    key, sizeof(struct brw_wm_prog_key),
                    program, program_size,
-                   &c->prog_data, sizeof(c->prog_data),
+                   &prog_data, sizeof(prog_data),
                    &brw->wm.base.prog_offset, &brw->wm.prog_data);
 
    ralloc_free(mem_ctx);
@@ -235,6 +268,14 @@ brw_debug_recompile_sampler_key(struct brw_context *brw,
                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
    found |= key_debug(brw, "gather channel quirk on any texture unit",
                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
+   found |= key_debug(brw, "compressed multisample layout",
+                      old_key->compressed_multisample_layout_mask,
+                      key->compressed_multisample_layout_mask);
+
+   for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
+      found |= key_debug(brw, "textureGather workarounds",
+                         old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
+   }
 
    return found;
 }
@@ -252,7 +293,7 @@ brw_wm_debug_recompile(struct brw_context *brw,
 
    for (unsigned int i = 0; i < brw->cache.size; i++) {
       for (c = brw->cache.items[i]; c; c = c->next) {
-         if (c->cache_id == BRW_WM_PROG) {
+         if (c->cache_id == BRW_CACHE_FS_PROG) {
             old_key = c->key;
 
             if (old_key->program_string_id == key->program_string_id)
@@ -275,6 +316,10 @@ brw_wm_debug_recompile(struct brw_context *brw,
                       old_key->stats_wm, key->stats_wm);
    found |= key_debug(brw, "flat shading",
                       old_key->flat_shade, key->flat_shade);
+   found |= key_debug(brw, "per-sample shading",
+                      old_key->persample_shading, key->persample_shading);
+   found |= key_debug(brw, "per-sample shading and 2x MSAA",
+                      old_key->persample_2x, key->persample_2x);
    found |= key_debug(brw, "number of color buffers",
                       old_key->nr_color_regions, key->nr_color_regions);
    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
@@ -347,7 +392,8 @@ brw_populate_sampler_prog_key_data(struct gl_context *ctx,
          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 
-        if (sampler->MinFilter != GL_NEAREST &&
+        if (brw->gen < 8 &&
+             sampler->MinFilter != GL_NEAREST &&
             sampler->MagFilter != GL_NEAREST) {
            if (sampler->WrapS == GL_CLAMP)
               key->gl_clamp_mask[0] |= 1 << s;
@@ -386,6 +432,27 @@ brw_populate_sampler_prog_key_data(struct gl_context *ctx,
    }
 }
 
+static bool
+brw_wm_state_dirty (struct brw_context *brw)
+{
+   return brw_state_dirty(brw,
+                          _NEW_BUFFERS |
+                          _NEW_COLOR |
+                          _NEW_DEPTH |
+                          _NEW_FRAG_CLAMP |
+                          _NEW_HINT |
+                          _NEW_LIGHT |
+                          _NEW_LINE |
+                          _NEW_MULTISAMPLE |
+                          _NEW_POLYGON |
+                          _NEW_STENCIL |
+                          _NEW_TEXTURE,
+                          BRW_NEW_FRAGMENT_PROGRAM |
+                          BRW_NEW_REDUCED_PRIMITIVE |
+                          BRW_NEW_STATS_WM |
+                          BRW_NEW_VUE_MAP_GEOM_OUT);
+}
+
 static void brw_wm_populate_key( struct brw_context *brw,
                                 struct brw_wm_prog_key *key )
 {
@@ -397,7 +464,7 @@ static void brw_wm_populate_key( struct brw_context *brw,
    GLuint lookup = 0;
    GLuint line_aa;
    bool program_uses_dfdy = fp->program.UsesDFdy;
-   bool multisample_fbo = ctx->DrawBuffer->Visual.samples > 1;
+   const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 
    memset(key, 0, sizeof(*key));
 
@@ -458,13 +525,8 @@ static void brw_wm_populate_key( struct brw_context *brw,
    key->line_aa = line_aa;
 
    /* _NEW_HINT */
-   if (brw->disable_derivative_optimization) {
-      key->high_quality_derivatives =
-         ctx->Hint.FragmentShaderDerivative != GL_FASTEST;
-   } else {
-      key->high_quality_derivatives =
-         ctx->Hint.FragmentShaderDerivative == GL_NICEST;
-   }
+   key->high_quality_derivatives =
+      ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 
    if (brw->gen < 6)
       key->stats_wm = brw->stats_wm;
@@ -501,7 +563,7 @@ static void brw_wm_populate_key( struct brw_context *brw,
     * drawable height in order to invert the Y axis.
     */
    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
-      key->drawable_height = ctx->DrawBuffer->Height;
+      key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
    }
 
    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
@@ -519,6 +581,8 @@ static void brw_wm_populate_key( struct brw_context *brw,
    /* Ignore sample qualifier while computing this flag. */
    key->persample_shading =
       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
+   if (key->persample_shading)
+      key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
 
    key->compute_pos_offset =
       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
@@ -550,47 +614,26 @@ static void brw_wm_populate_key( struct brw_context *brw,
    key->program_string_id = fp->id;
 }
 
-
-static void
+void
 brw_upload_wm_prog(struct brw_context *brw)
 {
    struct gl_context *ctx = &brw->ctx;
+   struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
    struct brw_wm_prog_key key;
    struct brw_fragment_program *fp = (struct brw_fragment_program *)
       brw->fragment_program;
 
+   if (!brw_wm_state_dirty(brw))
+      return;
+
    brw_wm_populate_key(brw, &key);
 
-   if (!brw_search_cache(&brw->cache, BRW_WM_PROG,
+   if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
                         &key, sizeof(key),
                         &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
-      bool success = do_wm_prog(brw, ctx->_Shader->_CurrentFragmentProgram, fp,
-                               &key);
+      bool success = brw_codegen_wm_prog(brw, current, fp, &key);
       (void) success;
       assert(success);
    }
    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 }
-
-
-const struct brw_tracked_state brw_wm_prog = {
-   .dirty = {
-      .mesa  = (_NEW_COLOR |
-               _NEW_DEPTH |
-               _NEW_STENCIL |
-               _NEW_POLYGON |
-               _NEW_LINE |
-               _NEW_HINT |
-               _NEW_LIGHT |
-               _NEW_FRAG_CLAMP |
-               _NEW_BUFFERS |
-               _NEW_TEXTURE |
-               _NEW_MULTISAMPLE),
-      .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
-               BRW_NEW_REDUCED_PRIMITIVE |
-                BRW_NEW_VUE_MAP_GEOM_OUT |
-                BRW_NEW_STATS_WM)
-   },
-   .emit = brw_upload_wm_prog
-};
-