st/nir: run st_nir_opts after 64bit ops lowering
[mesa.git] / src / mesa / state_tracker / st_cb_drawpixels.c
index 44f5b235c83ba0d786418a9b66a8cad4ca04797a..5f2d50ceb1b87e022ca3dc96a65630741b3dc472 100644 (file)
@@ -30,6 +30,7 @@
   *   Brian Paul
   */
 
+#include "main/errors.h"
 #include "main/imports.h"
 #include "main/image.h"
 #include "main/bufferobj.h"
@@ -64,6 +65,8 @@
 #include "st_sampler_view.h"
 #include "st_scissor.h"
 #include "st_texture.h"
+#include "st_util.h"
+#include "st_nir.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "util/u_format.h"
 #include "util/u_inlines.h"
 #include "util/u_math.h"
+#include "util/u_simple_shaders.h"
 #include "util/u_tile.h"
 #include "cso_cache/cso_context.h"
 
+#include "compiler/nir/nir_builder.h"
 
 /**
  * We have a simple glDrawPixels cache to try to optimize the case where the
  */
 #define USE_DRAWPIXELS_CACHE 1
 
+static nir_ssa_def *
+sample_via_nir(nir_builder *b, nir_variable *texcoord,
+               const char *name, int sampler)
+{
+   const struct glsl_type *sampler2D =
+      glsl_sampler_type(GLSL_SAMPLER_DIM_2D, false, false, GLSL_TYPE_FLOAT);
+
+   nir_variable *var =
+      nir_variable_create(b->shader, nir_var_uniform, sampler2D, name);
+   var->data.binding = sampler;
+   var->data.explicit_binding = true;
+
+   nir_deref_instr *deref = nir_build_deref_var(b, var);
+
+   nir_tex_instr *tex = nir_tex_instr_create(b->shader, 3);
+   tex->op = nir_texop_tex;
+   tex->sampler_dim = GLSL_SAMPLER_DIM_2D;
+   tex->coord_components = 2;
+   tex->dest_type = nir_type_float;
+   tex->src[0].src_type = nir_tex_src_texture_deref;
+   tex->src[0].src = nir_src_for_ssa(&deref->dest.ssa);
+   tex->src[1].src_type = nir_tex_src_sampler_deref;
+   tex->src[1].src = nir_src_for_ssa(&deref->dest.ssa);
+   tex->src[2].src_type = nir_tex_src_coord;
+   tex->src[2].src =
+      nir_src_for_ssa(nir_channels(b, nir_load_var(b, texcoord),
+                                   (1 << tex->coord_components) - 1));
+
+   nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, NULL);
+   nir_builder_instr_insert(b, &tex->instr);
+   return nir_channel(b, &tex->dest.ssa, 0);
+}
+
+static void *
+make_drawpix_z_stencil_program_nir(struct st_context *st,
+                                   bool write_depth,
+                                   bool write_stencil)
+{
+   struct nir_builder b;
+   const nir_shader_compiler_options *options =
+      st->ctx->Const.ShaderCompilerOptions[MESA_SHADER_FRAGMENT].NirOptions;
+
+   nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, options);
+
+   nir_variable *texcoord =
+      nir_variable_create(b.shader, nir_var_shader_in, glsl_vec_type(2),
+                          "texcoord");
+   texcoord->data.location = VARYING_SLOT_TEX0;
+
+   if (write_depth) {
+      nir_variable *out =
+         nir_variable_create(b.shader, nir_var_shader_out, glsl_float_type(),
+                             "gl_FragDepth");
+      out->data.location = FRAG_RESULT_DEPTH;
+      nir_ssa_def *depth = sample_via_nir(&b, texcoord, "depth", 0);
+      nir_store_var(&b, out, depth, 0x1);
+
+      /* Also copy color */
+      nir_variable *color_in =
+         nir_variable_create(b.shader, nir_var_shader_in, glsl_vec_type(4),
+                             "v_color");
+      color_in->data.location = VARYING_SLOT_COL0;
+
+      nir_variable *color_out =
+         nir_variable_create(b.shader, nir_var_shader_out, glsl_vec_type(4),
+                             "gl_FragColor");
+      color_out->data.location = FRAG_RESULT_COLOR;
+      nir_copy_var(&b, color_out, color_in);
+   }
+
+   if (write_stencil) {
+      nir_variable *out =
+         nir_variable_create(b.shader, nir_var_shader_out, glsl_uint_type(),
+                             "gl_FragStencilRefARB");
+      out->data.location = FRAG_RESULT_STENCIL;
+      nir_ssa_def *stencil = sample_via_nir(&b, texcoord, "stencil", 1);
+      nir_store_var(&b, out, stencil, 0x1);
+   }
+
+   char name[14];
+   snprintf(name, 14, "drawpixels %s%s",
+            write_depth ? "Z" : "", write_stencil ? "S" : "");
+
+   return st_nir_finish_builtin_shader(st, b.shader, name);
+}
 
 
-/**
- * Create fragment program that does a TEX() instruction to get a Z and/or
- * stencil value value, then writes to FRAG_RESULT_DEPTH/FRAG_RESULT_STENCIL.
- * Used for glDrawPixels(GL_DEPTH_COMPONENT / GL_STENCIL_INDEX).
- * Pass fragment color through as-is.
- *
- * \return CSO of the fragment shader.
- */
 static void *
-get_drawpix_z_stencil_program(struct st_context *st,
-                              GLboolean write_depth,
-                              GLboolean write_stencil)
+make_drawpix_z_stencil_program_tgsi(struct st_context *st,
+                                    bool write_depth,
+                                    bool write_stencil)
 {
    struct ureg_program *ureg;
    struct ureg_src depth_sampler, stencil_sampler;
    struct ureg_src texcoord, color;
    struct ureg_dst out_color, out_depth, out_stencil;
-   const GLuint shaderIndex = write_depth * 2 + write_stencil;
-   void *cso;
-
-   assert(shaderIndex < ARRAY_SIZE(st->drawpix.zs_shaders));
-
-   if (st->drawpix.zs_shaders[shaderIndex]) {
-      /* already have the proper shader */
-      return st->drawpix.zs_shaders[shaderIndex];
-   }
 
    ureg = ureg_create(PIPE_SHADER_FRAGMENT);
    if (ureg == NULL)
@@ -180,7 +253,42 @@ get_drawpix_z_stencil_program(struct st_context *st,
                TGSI_TEXTURE_2D, texcoord, stencil_sampler);
 
    ureg_END(ureg);
-   cso = ureg_create_shader_and_destroy(ureg, st->pipe);
+   return ureg_create_shader_and_destroy(ureg, st->pipe);
+}
+
+
+/**
+ * Create fragment program that does a TEX() instruction to get a Z and/or
+ * stencil value value, then writes to FRAG_RESULT_DEPTH/FRAG_RESULT_STENCIL.
+ * Used for glDrawPixels(GL_DEPTH_COMPONENT / GL_STENCIL_INDEX).
+ * Pass fragment color through as-is.
+ *
+ * \return CSO of the fragment shader.
+ */
+static void *
+get_drawpix_z_stencil_program(struct st_context *st,
+                              bool write_depth,
+                              bool write_stencil)
+{
+   struct pipe_screen *pscreen = st->pipe->screen;
+   const GLuint shaderIndex = write_depth * 2 + write_stencil;
+   void *cso;
+
+   assert(shaderIndex < ARRAY_SIZE(st->drawpix.zs_shaders));
+
+   if (st->drawpix.zs_shaders[shaderIndex]) {
+      /* already have the proper shader */
+      return st->drawpix.zs_shaders[shaderIndex];
+   }
+
+   enum pipe_shader_ir preferred_ir =
+      pscreen->get_shader_param(pscreen, PIPE_SHADER_FRAGMENT,
+                                PIPE_SHADER_CAP_PREFERRED_IR);
+
+   if (preferred_ir == PIPE_SHADER_IR_NIR)
+      cso = make_drawpix_z_stencil_program_nir(st, write_depth, write_stencil);
+   else
+      cso = make_drawpix_z_stencil_program_tgsi(st, write_depth, write_stencil);
 
    /* save the new shader */
    st->drawpix.zs_shaders[shaderIndex] = cso;
@@ -190,45 +298,42 @@ get_drawpix_z_stencil_program(struct st_context *st,
 
 /**
  * Create a simple vertex shader that just passes through the
- * vertex position and texcoord (and optionally, color).
+ * vertex position, texcoord, and color.
  */
-static void *
-make_passthrough_vertex_shader(struct st_context *st, 
-                               GLboolean passColor)
+void
+st_make_passthrough_vertex_shader(struct st_context *st)
 {
-   const unsigned texcoord_semantic = st->needs_texcoord_semantic ?
-      TGSI_SEMANTIC_TEXCOORD : TGSI_SEMANTIC_GENERIC;
-
-   if (!st->drawpix.vert_shaders[passColor]) {
-      struct ureg_program *ureg = ureg_create( PIPE_SHADER_VERTEX );
+   struct pipe_context *pipe = st->pipe;
+   struct pipe_screen *screen = pipe->screen;
 
-      if (ureg == NULL)
-         return NULL;
+   if (st->passthrough_vs)
+      return;
 
-      /* MOV result.pos, vertex.pos; */
-      ureg_MOV(ureg,
-               ureg_DECL_output( ureg, TGSI_SEMANTIC_POSITION, 0 ),
-               ureg_DECL_vs_input( ureg, 0 ));
+   enum pipe_shader_ir preferred_ir =
+      screen->get_shader_param(screen, PIPE_SHADER_VERTEX,
+                               PIPE_SHADER_CAP_PREFERRED_IR);
 
-      if (passColor) {
-         /* MOV result.color0, vertex.attr[1]; */
-         ureg_MOV(ureg,
-                  ureg_DECL_output( ureg, TGSI_SEMANTIC_COLOR, 0 ),
-                  ureg_DECL_vs_input( ureg, 1 ));
-      }
+   if (preferred_ir == PIPE_SHADER_IR_NIR) {
+      unsigned inputs[] =
+         {  VERT_ATTRIB_POS, VERT_ATTRIB_COLOR0, VERT_ATTRIB_GENERIC0 };
+      unsigned outputs[] =
+         { VARYING_SLOT_POS,  VARYING_SLOT_COL0,    VARYING_SLOT_TEX0 };
 
-      /* MOV result.texcoord0, vertex.attr[2]; */
-      ureg_MOV(ureg,
-               ureg_DECL_output( ureg, texcoord_semantic, 0 ),
-               ureg_DECL_vs_input( ureg, 2 ));
-
-      ureg_END( ureg );
-      
-      st->drawpix.vert_shaders[passColor] = 
-         ureg_create_shader_and_destroy( ureg, st->pipe );
+      st->passthrough_vs =
+         st_nir_make_passthrough_shader(st, "drawpixels VS",
+                                        MESA_SHADER_VERTEX, 3,
+                                        inputs, outputs, NULL, 0);
+   } else {
+      const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
+                                      TGSI_SEMANTIC_COLOR,
+        st->needs_texcoord_semantic ? TGSI_SEMANTIC_TEXCOORD :
+                                      TGSI_SEMANTIC_GENERIC };
+      const uint semantic_indexes[] = { 0, 0, 0 };
+
+      st->passthrough_vs =
+         util_make_vertex_passthrough_shader(st->pipe, 3, semantic_names,
+                                             semantic_indexes, false);
    }
-
-   return st->drawpix.vert_shaders[passColor];
 }
 
 
@@ -534,7 +639,7 @@ make_texture(struct st_context *st,
       GLenum intFormat = internal_format(ctx, format, type);
 
       pipeFormat = st_choose_format(st, intFormat, format, type,
-                                    st->internal_target, 0,
+                                    st->internal_target, 0, 0,
                                     PIPE_BIND_SAMPLER_VIEW, FALSE);
       assert(pipeFormat != PIPE_FORMAT_NONE);
    }
@@ -565,7 +670,11 @@ make_texture(struct st_context *st,
       dest = pipe_transfer_map(pipe, pt, 0, 0,
                                PIPE_TRANSFER_WRITE, 0, 0,
                                width, height, &transfer);
-
+      if (!dest) {
+         pipe_resource_reference(&pt, NULL);
+         _mesa_unmap_pbo_source(ctx, unpack);
+         return NULL;
+      }
 
       /* Put image into texture transfer.
        * Note that the image is actually going to be upside down in
@@ -676,7 +785,8 @@ draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
                                         ctx->Color._ClampFragmentColor;
       rasterizer.half_pixel_center = 1;
       rasterizer.bottom_edge_rule = 1;
-      rasterizer.depth_clip = !ctx->Transform.DepthClamp;
+      rasterizer.depth_clip_near = !ctx->Transform.DepthClampNear;
+      rasterizer.depth_clip_far = !ctx->Transform.DepthClampFar;
       rasterizer.scissor = ctx->Scissor.EnableFlags;
       cso_set_rasterizer(cso, &rasterizer);
    }
@@ -1129,7 +1239,7 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
               GLenum format, GLenum type,
               const struct gl_pixelstore_attrib *unpack, const void *pixels)
 {
-   void *driver_vp, *driver_fp;
+   void *driver_fp;
    struct st_context *st = st_context(ctx);
    struct pipe_context *pipe = st->pipe;
    GLboolean write_stencil = GL_FALSE, write_depth = GL_FALSE;
@@ -1147,7 +1257,7 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
    st_flush_bitmap_cache(st);
    st_invalidate_readpix_cache(st);
 
-   st_validate_state(st, ST_PIPELINE_RENDER);
+   st_validate_state(st, ST_PIPELINE_META);
 
    /* Limit the size of the glDrawPixels to the max texture size.
     * Strictly speaking, that's not correct but since we don't handle
@@ -1172,19 +1282,26 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
       return;
    }
 
+   /* Put glDrawPixels image into a texture */
+   pt = make_texture(st, width, height, format, type, unpack, pixels);
+   if (!pt) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glDrawPixels");
+      return;
+   }
+
+   st_make_passthrough_vertex_shader(st);
+
    /*
     * Get vertex/fragment shaders
     */
    if (write_depth || write_stencil) {
       driver_fp = get_drawpix_z_stencil_program(st, write_depth,
                                                 write_stencil);
-      driver_vp = make_passthrough_vertex_shader(st, GL_TRUE);
    }
    else {
       fpv = get_color_fp_variant(st);
 
       driver_fp = fpv->driver_shader;
-      driver_vp = make_passthrough_vertex_shader(st, GL_FALSE);
 
       if (ctx->Pixel.MapColorFlag) {
          pipe_sampler_view_reference(&sv[1],
@@ -1198,13 +1315,6 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
       st_upload_constants(st, &st->fp->Base);
    }
 
-   /* Put glDrawPixels image into a texture */
-   pt = make_texture(st, width, height, format, type, unpack, pixels);
-   if (!pt) {
-      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glDrawPixels");
-      return;
-   }
-
    /* create sampler view for the image */
    sv[0] = st_create_texture_sampler_view(st->pipe, pt);
    if (!sv[0]) {
@@ -1240,7 +1350,7 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
                       ctx->Pixel.ZoomX, ctx->Pixel.ZoomY,
                       sv,
                       num_sampler_view,
-                      driver_vp,
+                      st->passthrough_vs,
                       driver_fp, fpv,
                       ctx->Current.RasterColor,
                       GL_FALSE, write_depth, write_stencil);
@@ -1474,10 +1584,12 @@ blit_copy_pixels(struct gl_context *ctx, GLint srcx, GLint srcy,
          if (screen->is_format_supported(screen, blit.src.format,
                                          blit.src.resource->target,
                                          blit.src.resource->nr_samples,
+                                         blit.src.resource->nr_storage_samples,
                                          PIPE_BIND_SAMPLER_VIEW) &&
              screen->is_format_supported(screen, blit.dst.format,
                                          blit.dst.resource->target,
                                          blit.dst.resource->nr_samples,
+                                         blit.dst.resource->nr_storage_samples,
                                          PIPE_BIND_RENDER_TARGET)) {
             pipe->blit(pipe, &blit);
             return GL_TRUE;
@@ -1498,7 +1610,7 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
    struct pipe_context *pipe = st->pipe;
    struct pipe_screen *screen = pipe->screen;
    struct st_renderbuffer *rbRead;
-   void *driver_vp, *driver_fp;
+   void *driver_fp;
    struct pipe_resource *pt;
    struct pipe_sampler_view *sv[2] = { NULL };
    struct st_fp_variant *fpv = NULL;
@@ -1514,7 +1626,7 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
    st_flush_bitmap_cache(st);
    st_invalidate_readpix_cache(st);
 
-   st_validate_state(st, ST_PIPELINE_RENDER);
+   st_validate_state(st, ST_PIPELINE_META);
 
    if (type == GL_DEPTH_STENCIL) {
       /* XXX make this more efficient */
@@ -1539,6 +1651,7 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
     * are handled.
     */
 
+   st_make_passthrough_vertex_shader(st);
 
    /*
     * Get vertex/fragment shaders
@@ -1549,7 +1662,6 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
       rbRead = st_get_color_read_renderbuffer(ctx);
 
       driver_fp = fpv->driver_shader;
-      driver_vp = make_passthrough_vertex_shader(st, GL_FALSE);
 
       if (ctx->Pixel.MapColorFlag) {
          pipe_sampler_view_reference(&sv[1],
@@ -1568,7 +1680,6 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
                                Attachment[BUFFER_DEPTH].Renderbuffer);
 
       driver_fp = get_drawpix_z_stencil_program(st, GL_TRUE, GL_FALSE);
-      driver_vp = make_passthrough_vertex_shader(st, GL_TRUE);
    }
 
    /* Choose the format for the temporary texture. */
@@ -1577,11 +1688,11 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
       (type == GL_COLOR ? PIPE_BIND_RENDER_TARGET : PIPE_BIND_DEPTH_STENCIL);
 
    if (!screen->is_format_supported(screen, srcFormat, st->internal_target, 0,
-                                    srcBind)) {
+                                    0, srcBind)) {
       /* srcFormat is non-renderable. Find a compatible renderable format. */
       if (type == GL_DEPTH) {
          srcFormat = st_choose_format(st, GL_DEPTH_COMPONENT, GL_NONE,
-                                      GL_NONE, st->internal_target, 0,
+                                      GL_NONE, st->internal_target, 0, 0,
                                       srcBind, FALSE);
       }
       else {
@@ -1589,27 +1700,27 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
 
          if (util_format_is_float(srcFormat)) {
             srcFormat = st_choose_format(st, GL_RGBA32F, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
+                                         GL_NONE, st->internal_target, 0, 0,
                                          srcBind, FALSE);
          }
          else if (util_format_is_pure_sint(srcFormat)) {
             srcFormat = st_choose_format(st, GL_RGBA32I, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
+                                         GL_NONE, st->internal_target, 0, 0,
                                          srcBind, FALSE);
          }
          else if (util_format_is_pure_uint(srcFormat)) {
             srcFormat = st_choose_format(st, GL_RGBA32UI, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
+                                         GL_NONE, st->internal_target, 0, 0,
                                          srcBind, FALSE);
          }
          else if (util_format_is_snorm(srcFormat)) {
             srcFormat = st_choose_format(st, GL_RGBA16_SNORM, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
+                                         GL_NONE, st->internal_target, 0, 0,
                                          srcBind, FALSE);
          }
          else {
             srcFormat = st_choose_format(st, GL_RGBA, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
+                                         GL_NONE, st->internal_target, 0, 0,
                                          srcBind, FALSE);
          }
       }
@@ -1695,7 +1806,7 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
                       width, height, ctx->Pixel.ZoomX, ctx->Pixel.ZoomY,
                       sv,
                       num_sampler_view,
-                      driver_vp, 
+                      st->passthrough_vs,
                       driver_fp, fpv,
                       ctx->Current.Attrib[VERT_ATTRIB_COLOR0],
                       invertTex, GL_FALSE, GL_FALSE);
@@ -1724,10 +1835,8 @@ st_destroy_drawpix(struct st_context *st)
                                     st->drawpix.zs_shaders[i]);
    }
 
-   if (st->drawpix.vert_shaders[0])
-      cso_delete_vertex_shader(st->cso_context, st->drawpix.vert_shaders[0]);
-   if (st->drawpix.vert_shaders[1])
-      cso_delete_vertex_shader(st->cso_context, st->drawpix.vert_shaders[1]);
+   if (st->passthrough_vs)
+      cso_delete_vertex_shader(st->cso_context, st->passthrough_vs);
 
    /* Free cache data */
    for (i = 0; i < ARRAY_SIZE(st->drawpix_cache.entries); i++) {