gallium: add nir lowering passes for the draw pipe stages. (v2)
authorDave Airlie <airlied@redhat.com>
Thu, 5 Sep 2019 05:47:39 +0000 (15:47 +1000)
committerDave Airlie <airlied@redhat.com>
Thu, 28 Nov 2019 04:49:05 +0000 (14:49 +1000)
This transforms the NIR shaders like the TGSI transforms worked.

v2: fix some nir info requirements, use 32-bit bools

Acked-by: Roland Scheidegger <sroland@vmware.com>
src/gallium/auxiliary/Makefile.sources
src/gallium/auxiliary/draw/draw_pipe_aaline.c
src/gallium/auxiliary/draw/draw_pipe_aapoint.c
src/gallium/auxiliary/draw/draw_pipe_pstipple.c
src/gallium/auxiliary/meson.build
src/gallium/auxiliary/nir/nir_draw_helpers.c [new file with mode: 0644]
src/gallium/auxiliary/nir/nir_draw_helpers.h [new file with mode: 0644]

index 4ee99b8f06367257a4547bde2e434e59aa29896d..438cb6009537f80c6a8eac3756e961bd6af96f66 100644 (file)
@@ -319,7 +319,9 @@ NIR_SOURCES := \
        nir/tgsi_to_nir.c \
        nir/tgsi_to_nir.h \
        nir/nir_to_tgsi_info.c \
-       nir/nir_to_tgsi_info.h
+       nir/nir_to_tgsi_info.h \
+       nir/nir_draw_helpers.c \
+       nir/nir_draw_helpers.h
 
 VL_SOURCES := \
        vl/vl_bicubic_filter.c \
index e84fb5d615765c84dbea1ed99a514b9995c61f1c..872b29e9c8da76c1b315f49742f8683b2204063e 100644 (file)
@@ -48,6 +48,8 @@
 #include "draw_private.h"
 #include "draw_pipe.h"
 
+#include "nir.h"
+#include "nir/nir_draw_helpers.h"
 
 /** Approx number of new tokens for instructions in aa_transform_inst() */
 #define NUM_NEW_TOKENS 53
@@ -318,6 +320,30 @@ fail:
    return FALSE;
 }
 
+static boolean
+generate_aaline_fs_nir(struct aaline_stage *aaline)
+{
+#ifdef LLVM_AVAILABLE
+   struct pipe_context *pipe = aaline->stage.draw->pipe;
+   const struct pipe_shader_state *orig_fs = &aaline->fs->state;
+   struct pipe_shader_state aaline_fs;
+
+   aaline_fs = *orig_fs; /* copy to init */
+   aaline_fs.ir.nir = nir_shader_clone(NULL, orig_fs->ir.nir);
+   if (!aaline_fs.ir.nir)
+      return FALSE;
+
+   nir_lower_aaline_fs(aaline_fs.ir.nir, &aaline->fs->generic_attrib);
+   aaline->fs->aaline_fs = aaline->driver_create_fs_state(pipe, &aaline_fs);
+   if (aaline->fs->aaline_fs == NULL)
+      goto fail;
+
+   return TRUE;
+
+fail:
+#endif
+   return FALSE;
+}
 
 /**
  * When we're about to draw our first AA line in a batch, this function is
@@ -329,8 +355,14 @@ bind_aaline_fragment_shader(struct aaline_stage *aaline)
    struct draw_context *draw = aaline->stage.draw;
    struct pipe_context *pipe = draw->pipe;
 
-   if (!aaline->fs->aaline_fs && !generate_aaline_fs(aaline))
-      return FALSE;
+   if (!aaline->fs->aaline_fs) {
+      if (aaline->fs->state.type == PIPE_SHADER_IR_NIR) {
+         if (!generate_aaline_fs_nir(aaline))
+            return FALSE;
+      } else
+         if (!generate_aaline_fs(aaline))
+            return FALSE;
+   }
 
    draw->suspend_flushing = TRUE;
    aaline->driver_bind_fs_state(pipe, aaline->fs->aaline_fs);
@@ -618,7 +650,13 @@ aaline_create_fs_state(struct pipe_context *pipe,
    if (!aafs)
       return NULL;
 
-   aafs->state.tokens = tgsi_dup_tokens(fs->tokens);
+   aafs->state.type = fs->type;
+   if (fs->type == PIPE_SHADER_IR_TGSI)
+      aafs->state.tokens = tgsi_dup_tokens(fs->tokens);
+#ifdef LLVM_AVAILABLE
+   else
+      aafs->state.ir.nir = nir_shader_clone(NULL, fs->ir.nir);
+#endif
 
    /* pass-through */
    aafs->driver_fs = aaline->driver_create_fs_state(pipe, fs);
@@ -662,7 +700,10 @@ aaline_delete_fs_state(struct pipe_context *pipe, void *fs)
          aaline->driver_delete_fs_state(pipe, aafs->aaline_fs);
    }
 
-   FREE((void*)aafs->state.tokens);
+   if (aafs->state.type == PIPE_SHADER_IR_TGSI)
+      FREE((void*)aafs->state.tokens);
+   else
+      ralloc_free(aafs->state.ir.nir);
    FREE(aafs);
 }
 
@@ -681,9 +722,12 @@ draw_aaline_prepare_outputs(struct draw_context *draw,
       return;
 
    /* allocate the extra post-transformed vertex attribute */
-   aaline->coord_slot = draw_alloc_extra_vertex_attrib(draw,
-                                                     TGSI_SEMANTIC_GENERIC,
-                                                     aaline->fs->generic_attrib);
+   if (aaline->fs->aaline_fs)
+      aaline->coord_slot = draw_alloc_extra_vertex_attrib(draw,
+                                                          TGSI_SEMANTIC_GENERIC,
+                                                          aaline->fs->generic_attrib);
+   else
+      aaline->coord_slot = -1;
 }
 
 /**
index dc22039b127c84a570c6d7dc3de280d50b529495..797e0d2d9c2be588a9c8abc7dd8c345cabfd2fb0 100644 (file)
@@ -52,6 +52,8 @@
 #include "draw_vs.h"
 #include "draw_pipe.h"
 
+#include "nir.h"
+#include "nir/nir_draw_helpers.h"
 
 /** Approx number of new tokens for instructions in aa_transform_inst() */
 #define NUM_NEW_TOKENS 200
@@ -364,6 +366,8 @@ generate_aapoint_fs(struct aapoint_stage *aapoint)
    struct pipe_context *pipe = aapoint->stage.draw->pipe;
 
    aapoint_fs = *orig_fs; /* copy to init */
+
+   assert(aapoint_fs.type == PIPE_SHADER_IR_TGSI);
    aapoint_fs.tokens = tgsi_alloc_tokens(newLen);
    if (aapoint_fs.tokens == NULL)
       return FALSE;
@@ -404,6 +408,30 @@ fail:
    return FALSE;
 }
 
+static boolean
+generate_aapoint_fs_nir(struct aapoint_stage *aapoint)
+{
+#ifdef LLVM_AVAILABLE
+   struct pipe_context *pipe = aapoint->stage.draw->pipe;
+   const struct pipe_shader_state *orig_fs = &aapoint->fs->state;
+   struct pipe_shader_state aapoint_fs;
+
+   aapoint_fs = *orig_fs; /* copy to init */
+   aapoint_fs.ir.nir = nir_shader_clone(NULL, orig_fs->ir.nir);
+   if (!aapoint_fs.ir.nir)
+      return FALSE;
+
+   nir_lower_aapoint_fs(aapoint_fs.ir.nir, &aapoint->fs->generic_attrib);
+   aapoint->fs->aapoint_fs = aapoint->driver_create_fs_state(pipe, &aapoint_fs);
+   if (aapoint->fs->aapoint_fs == NULL)
+      goto fail;
+
+   return TRUE;
+
+fail:
+#endif
+   return FALSE;
+}
 
 /**
  * When we're about to draw our first AA point in a batch, this function is
@@ -415,9 +443,13 @@ bind_aapoint_fragment_shader(struct aapoint_stage *aapoint)
    struct draw_context *draw = aapoint->stage.draw;
    struct pipe_context *pipe = draw->pipe;
 
-   if (!aapoint->fs->aapoint_fs &&
-       !generate_aapoint_fs(aapoint))
-      return FALSE;
+   if (!aapoint->fs->aapoint_fs) {
+      if (aapoint->fs->state.type == PIPE_SHADER_IR_NIR) {
+         if (!generate_aapoint_fs_nir(aapoint))
+            return FALSE;
+      } else if (!generate_aapoint_fs(aapoint))
+         return FALSE;
+   }
 
    draw->suspend_flushing = TRUE;
    aapoint->driver_bind_fs_state(pipe, aapoint->fs->aapoint_fs);
@@ -637,11 +669,14 @@ draw_aapoint_prepare_outputs(struct draw_context *draw,
    if (!rast->point_smooth)
       return;
 
-   /* allocate the extra post-transformed vertex attribute */
-   aapoint->tex_slot = draw_alloc_extra_vertex_attrib(draw,
-                                                      TGSI_SEMANTIC_GENERIC,
-                                                      aapoint->fs->generic_attrib);
-   assert(aapoint->tex_slot > 0); /* output[0] is vertex pos */
+   if (aapoint->fs->aapoint_fs) {
+      /* allocate the extra post-transformed vertex attribute */
+      aapoint->tex_slot = draw_alloc_extra_vertex_attrib(draw,
+                                                         TGSI_SEMANTIC_GENERIC,
+                                                         aapoint->fs->generic_attrib);
+      assert(aapoint->tex_slot > 0); /* output[0] is vertex pos */
+   } else
+      aapoint->tex_slot = -1;
 
    /* find psize slot in post-transform vertex */
    aapoint->psize_slot = -1;
@@ -710,8 +745,13 @@ aapoint_create_fs_state(struct pipe_context *pipe,
    if (!aafs)
       return NULL;
 
-   aafs->state.tokens = tgsi_dup_tokens(fs->tokens);
-
+   aafs->state.type = fs->type;
+   if (fs->type == PIPE_SHADER_IR_TGSI)
+      aafs->state.tokens = tgsi_dup_tokens(fs->tokens);
+#ifdef LLVM_AVAILABLE
+   else
+      aafs->state.ir.nir = nir_shader_clone(NULL, fs->ir.nir);
+#endif
    /* pass-through */
    aafs->driver_fs = aapoint->driver_create_fs_state(pipe, fs);
 
@@ -744,7 +784,10 @@ aapoint_delete_fs_state(struct pipe_context *pipe, void *fs)
    if (aafs->aapoint_fs)
       aapoint->driver_delete_fs_state(pipe, aafs->aapoint_fs);
 
-   FREE((void*)aafs->state.tokens);
+   if (aafs->state.type == PIPE_SHADER_IR_TGSI)
+      FREE((void*)aafs->state.tokens);
+   else
+      ralloc_free(aafs->state.ir.nir);
 
    FREE(aafs);
 }
index bd6637a6730ce827b39e54e2647e44a40e1e587e..bd74cb42adaebac368d96232db07929e17a099d6 100644 (file)
@@ -51,6 +51,8 @@
 #include "draw_context.h"
 #include "draw_pipe.h"
 
+#include "nir.h"
+#include "nir/nir_draw_helpers.h"
 
 /** Approx number of new tokens for instructions in pstip_transform_inst() */
 #define NUM_NEW_TOKENS 53
@@ -133,12 +135,20 @@ generate_pstip_fs(struct pstip_stage *pstip)
                    TGSI_FILE_SYSTEM_VALUE : TGSI_FILE_INPUT;
 
    pstip_fs = *orig_fs; /* copy to init */
-   pstip_fs.tokens = util_pstipple_create_fragment_shader(orig_fs->tokens,
-                                                          &pstip->fs->sampler_unit,
-                                                          0,
-                                                          wincoord_file);
-   if (pstip_fs.tokens == NULL)
-      return FALSE;
+   if (orig_fs->type == PIPE_SHADER_IR_TGSI) {
+      pstip_fs.tokens = util_pstipple_create_fragment_shader(orig_fs->tokens,
+                                                             &pstip->fs->sampler_unit,
+                                                             0,
+                                                             wincoord_file);
+      if (pstip_fs.tokens == NULL)
+         return FALSE;
+   } else {
+#ifdef LLVM_AVAILABLE
+      pstip_fs.ir.nir = nir_shader_clone(NULL, orig_fs->ir.nir);
+      nir_lower_pstipple_fs(pstip_fs.ir.nir,
+                            &pstip->fs->sampler_unit, 0, wincoord_file == TGSI_FILE_SYSTEM_VALUE);
+#endif
+   }
 
    assert(pstip->fs->sampler_unit < PIPE_MAX_SAMPLERS);
 
@@ -334,7 +344,11 @@ pstip_create_fs_state(struct pipe_context *pipe,
    struct pstip_fragment_shader *pstipfs = CALLOC_STRUCT(pstip_fragment_shader);
 
    if (pstipfs) {
-      pstipfs->state.tokens = tgsi_dup_tokens(fs->tokens);
+      pstipfs->state.type = fs->type;
+      if (fs->type == PIPE_SHADER_IR_TGSI)
+         pstipfs->state.tokens = tgsi_dup_tokens(fs->tokens);
+      else
+         pstipfs->state.ir.nir = nir_shader_clone(NULL, fs->ir.nir);
 
       /* pass-through */
       pstipfs->driver_fs = pstip->driver_create_fs_state(pstip->pipe, fs);
@@ -368,7 +382,10 @@ pstip_delete_fs_state(struct pipe_context *pipe, void *fs)
    if (pstipfs->pstip_fs)
       pstip->driver_delete_fs_state(pstip->pipe, pstipfs->pstip_fs);
 
-   FREE((void*)pstipfs->state.tokens);
+   if (pstipfs->state.type == PIPE_SHADER_IR_TGSI)
+      FREE((void*)pstipfs->state.tokens);
+   else
+      ralloc_free(pstipfs->state.ir.nir);
    FREE(pstipfs);
 }
 
index ebc5e5449e054f53089965fb85de5658679082e9..f96b32e9d752cc6bb69e77838efcecfec975c473 100644 (file)
@@ -336,6 +336,8 @@ files_libgallium = files(
   'util/u_viewport.h',
   'nir/tgsi_to_nir.c',
   'nir/tgsi_to_nir.h',
+  'nir/nir_draw_helpers.c',
+  'nir/nir_draw_helpers.h',
 )
 
 if dep_libdrm.found()
diff --git a/src/gallium/auxiliary/nir/nir_draw_helpers.c b/src/gallium/auxiliary/nir/nir_draw_helpers.c
new file mode 100644 (file)
index 0000000..036ce84
--- /dev/null
@@ -0,0 +1,372 @@
+/**************************************************************************
+ *
+ * Copyright 2019 Red Hat.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ **************************************************************************/
+
+/*
+ * NIR lowering passes to handle the draw stages for
+ * - pstipple
+ * - aaline
+ * - aapoint.
+ *
+ * These are all ported from the equivalent TGSI transforms.
+ */
+
+#include "nir.h"
+#include "tgsi/tgsi_from_mesa.h"
+#include "nir_builder.h"
+
+#include "nir_draw_helpers.h"
+
+typedef struct {
+   nir_builder b;
+   nir_shader *shader;
+   bool fs_pos_is_sysval;
+   nir_variable *stip_tex;
+   nir_ssa_def *fragcoord;
+} lower_pstipple;
+
+static nir_ssa_def *
+load_frag_coord(nir_builder *b)
+{
+   int max_driver_loc = -1;
+   nir_foreach_variable(var, &b->shader->inputs) {
+      if (var->data.location == VARYING_SLOT_POS)
+         return nir_load_var(b, var);
+      if (max_driver_loc < (int)var->data.driver_location)
+         max_driver_loc = var->data.driver_location;
+   }
+
+   nir_variable *pos = nir_variable_create(b->shader, nir_var_shader_in,
+                                           glsl_vec4_type(), NULL);
+   pos->data.location = VARYING_SLOT_POS;
+   pos->data.interpolation = INTERP_MODE_NOPERSPECTIVE;
+   pos->data.driver_location = max_driver_loc + 1;
+   b->shader->num_inputs++;
+   return nir_load_var(b, pos);
+}
+
+static void
+nir_lower_pstipple_block(nir_block *block,
+                         lower_pstipple *state)
+{
+   nir_builder *b = &state->b;
+   nir_ssa_def *texcoord;
+
+   b->cursor = nir_before_block(block);
+
+   nir_ssa_def *div32 = nir_imm_vec2(b, 1.0/32.0, 1.0/32.0);
+
+   nir_ssa_def *frag_coord = state->fs_pos_is_sysval ? nir_load_frag_coord(b) : load_frag_coord(b);
+
+   texcoord = nir_fmul(b, frag_coord, div32);
+
+   nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1);
+   tex->op = nir_texop_tex;
+   tex->sampler_dim = GLSL_SAMPLER_DIM_2D;
+   tex->coord_components = 2;
+   tex->dest_type = nir_type_float;
+   tex->texture_index = state->stip_tex->data.binding;
+   tex->sampler_index = state->stip_tex->data.binding;
+   tex->src[0].src_type = nir_tex_src_coord;
+   tex->src[0].src = nir_src_for_ssa(texcoord);
+   nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, NULL);
+
+   nir_builder_instr_insert(b, &tex->instr);
+
+   nir_ssa_def *condition = nir_f2b32(b, nir_channel(b, &tex->dest.ssa, 3));
+   nir_intrinsic_instr *discard = nir_intrinsic_instr_create(b->shader, nir_intrinsic_discard_if);
+   discard->src[0] = nir_src_for_ssa(condition);
+   nir_builder_instr_insert(b, &discard->instr);
+   b->shader->info.fs.uses_discard = true;
+}
+
+static void
+nir_lower_pstipple_impl(nir_function_impl *impl,
+                        lower_pstipple *state)
+{
+   nir_builder *b = &state->b;
+
+   nir_builder_init(b, impl);
+
+   nir_block *start = nir_start_block(impl);
+   nir_lower_pstipple_block(start, state);
+}
+
+void
+nir_lower_pstipple_fs(struct nir_shader *shader,
+                      unsigned *samplerUnitOut,
+                      unsigned fixedUnit,
+                      bool fs_pos_is_sysval)
+{
+   lower_pstipple state = {
+      .shader = shader,
+      .fs_pos_is_sysval = fs_pos_is_sysval,
+   };
+   if (shader->info.stage != MESA_SHADER_FRAGMENT)
+      return;
+
+   int binding = 0;
+   nir_foreach_variable(var, &shader->uniforms) {
+      if (glsl_type_is_sampler(var->type)) {
+         if (var->data.binding >= binding)
+            binding = var->data.binding + 1;
+      }
+   }
+   const struct glsl_type *sampler2D =
+      glsl_sampler_type(GLSL_SAMPLER_DIM_2D, false, false, GLSL_TYPE_FLOAT);
+
+   nir_variable *tex_var = nir_variable_create(shader, nir_var_uniform, sampler2D, "stipple_tex");
+   tex_var->data.binding = binding;
+   tex_var->data.explicit_binding = true;
+   tex_var->data.how_declared = nir_var_hidden;
+
+   shader->info.textures_used |= (1 << binding);
+   state.stip_tex = tex_var;
+
+   nir_foreach_function(function, shader) {
+      if (function->impl) {
+         nir_lower_pstipple_impl(function->impl, &state);
+      }
+   }
+   *samplerUnitOut = binding;
+}
+
+typedef struct {
+   nir_builder b;
+   nir_shader *shader;
+   nir_variable *line_width_input;
+} lower_aaline;
+
+static void
+nir_lower_aaline_block(nir_block *block,
+                       lower_aaline *state)
+{
+  nir_builder *b = &state->b;
+  nir_foreach_instr(instr, block) {
+      if (instr->type != nir_instr_type_intrinsic)
+         continue;
+
+      nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
+      if (intrin->intrinsic != nir_intrinsic_store_deref)
+         continue;
+
+      nir_variable *var = nir_intrinsic_get_var(intrin, 0);
+      if (var->data.mode != nir_var_shader_out)
+         continue;
+      if (var->data.location != FRAG_RESULT_COLOR)
+         continue;
+
+      nir_ssa_def *out_input = intrin->src[1].ssa;
+      b->cursor = nir_before_instr(instr);
+      nir_ssa_def *lw = nir_load_var(b, state->line_width_input);
+      nir_ssa_def *tmp = nir_fsat(b, nir_fadd(b, nir_channel(b, lw, 1),
+                                              nir_fneg(b, nir_fabs(b, nir_channel(b, lw, 0)))));
+      nir_ssa_def *tmp1 = nir_fsat(b, nir_fadd(b, nir_channel(b, lw, 3),
+                                               nir_fneg(b, nir_fabs(b, nir_channel(b, lw, 2)))));
+
+      tmp = nir_fmul(b, tmp, tmp1);
+      tmp = nir_fmul(b, nir_channel(b, out_input, 3), tmp);
+
+      nir_ssa_def *out = nir_vec4(b, nir_channel(b, out_input, 0),
+                                  nir_channel(b, out_input, 1),
+                                  nir_channel(b, out_input, 2),
+                                  tmp);
+      nir_instr_rewrite_src(instr, &intrin->src[1], nir_src_for_ssa(out));
+   }
+
+}
+
+static void
+nir_lower_aaline_impl(nir_function_impl *impl,
+                      lower_aaline *state)
+{
+   nir_builder *b = &state->b;
+
+   nir_builder_init(b, impl);
+
+   nir_foreach_block(block, impl) {
+      nir_lower_aaline_block(block, state);
+   }
+}
+
+void
+nir_lower_aaline_fs(struct nir_shader *shader, int *varying)
+{
+   lower_aaline state = {
+      .shader = shader,
+   };
+   if (shader->info.stage != MESA_SHADER_FRAGMENT)
+      return;
+
+   int highest_location = -1, highest_drv_location = -1;
+   nir_foreach_variable(var, &shader->inputs) {
+     if ((int)var->data.location > highest_location)
+         highest_location = var->data.location;
+     if ((int)var->data.driver_location > highest_drv_location)
+         highest_drv_location = var->data.driver_location;
+   }
+
+   nir_variable *line_width = nir_variable_create(shader, nir_var_shader_in,
+                                                  glsl_vec4_type(), "aaline");
+   if (highest_location == -1 || highest_location < VARYING_SLOT_VAR0) {
+     line_width->data.location = VARYING_SLOT_VAR0;
+     line_width->data.driver_location = highest_drv_location + 1;
+   } else {
+     line_width->data.location = highest_location + 1;
+     line_width->data.driver_location = highest_drv_location + 1;
+   }
+   shader->num_inputs++;
+   *varying = tgsi_get_generic_gl_varying_index(line_width->data.location, false);
+   state.line_width_input = line_width;
+
+   nir_foreach_function(function, shader) {
+      if (function->impl) {
+         nir_lower_aaline_impl(function->impl, &state);
+      }
+   }
+}
+
+typedef struct {
+   nir_builder b;
+   nir_shader *shader;
+   nir_variable *input;
+} lower_aapoint;
+
+static void
+nir_lower_aapoint_block(nir_block *block,
+                        lower_aapoint *state, nir_ssa_def *sel)
+{
+  nir_builder *b = &state->b;
+  nir_foreach_instr(instr, block) {
+      if (instr->type != nir_instr_type_intrinsic)
+         continue;
+
+      nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
+      if (intrin->intrinsic != nir_intrinsic_store_deref)
+         continue;
+
+      nir_variable *var = nir_intrinsic_get_var(intrin, 0);
+      if (var->data.mode != nir_var_shader_out)
+         continue;
+      if (var->data.location != FRAG_RESULT_COLOR)
+         continue;
+
+      nir_ssa_def *out_input = intrin->src[1].ssa;
+      b->cursor = nir_before_instr(instr);
+
+      nir_ssa_def *tmp = nir_fmul(b, nir_channel(b, out_input, 3), sel);
+      nir_ssa_def *out = nir_vec4(b, nir_channel(b, out_input, 0),
+                                  nir_channel(b, out_input, 1),
+                                  nir_channel(b, out_input, 2),
+                                  tmp);
+      nir_instr_rewrite_src(instr, &intrin->src[1], nir_src_for_ssa(out));
+   }
+
+}
+
+static void
+nir_lower_aapoint_impl(nir_function_impl *impl,
+                      lower_aapoint *state)
+{
+   nir_builder *b = &state->b;
+
+   nir_builder_init(b, impl);
+
+   nir_block *block = nir_start_block(impl);
+   b->cursor = nir_before_block(block);
+
+   nir_ssa_def *aainput = nir_load_var(b, state->input);
+
+   nir_ssa_def *dist = nir_fadd(b, nir_fmul(b, nir_channel(b, aainput, 0), nir_channel(b, aainput, 0)),
+                                nir_fmul(b, nir_channel(b, aainput, 1), nir_channel(b, aainput, 1)));
+
+   nir_ssa_def *k = nir_channel(b, aainput, 2);
+   nir_ssa_def *chan_val_one = nir_channel(b, aainput, 3);
+   nir_ssa_def *comp = nir_flt32(b, chan_val_one, dist);
+
+   nir_intrinsic_instr *discard = nir_intrinsic_instr_create(b->shader, nir_intrinsic_discard_if);
+   discard->src[0] = nir_src_for_ssa(comp);
+   nir_builder_instr_insert(b, &discard->instr);
+   b->shader->info.fs.uses_discard = true;
+
+   /* compute coverage factor = (1-d)/(1-k) */
+   /* 1 - k */
+   nir_ssa_def *tmp = nir_fadd(b, chan_val_one, nir_fneg(b, k));
+   /* 1.0 / (1 - k) */
+   tmp = nir_frcp(b, tmp);
+
+   /* 1 - d */
+   nir_ssa_def *tmp2 = nir_fadd(b, chan_val_one, nir_fneg(b, dist));
+
+   /* (1 - d) / (1 - k) */
+   nir_ssa_def *coverage = nir_fmul(b, tmp, tmp2);
+
+   /* if (k >= distance)
+    *    sel = coverage;
+    * else
+    *    sel = 1.0;
+    */
+   nir_ssa_def *sel = nir_b32csel(b, nir_fge32(b, k, dist), coverage, chan_val_one);
+
+   nir_foreach_block(block, impl) {
+     nir_lower_aapoint_block(block, state, sel);
+   }
+}
+
+void
+nir_lower_aapoint_fs(struct nir_shader *shader, int *varying)
+{
+   lower_aapoint state = {
+      .shader = shader,
+   };
+   if (shader->info.stage != MESA_SHADER_FRAGMENT)
+      return;
+
+   int highest_location = -1, highest_drv_location = -1;
+   nir_foreach_variable(var, &shader->inputs) {
+     if ((int)var->data.location > highest_location)
+         highest_location = var->data.location;
+     if ((int)var->data.driver_location > highest_drv_location)
+         highest_drv_location = var->data.driver_location;
+   }
+
+   nir_variable *aapoint_input = nir_variable_create(shader, nir_var_shader_in,
+                                                     glsl_vec4_type(), "aapoint");
+   if (highest_location == -1 || highest_location < VARYING_SLOT_VAR0) {
+     aapoint_input->data.location = VARYING_SLOT_VAR0;
+   } else {
+     aapoint_input->data.location = highest_location + 1;
+   }
+   aapoint_input->data.driver_location = highest_drv_location + 1;
+
+   shader->num_inputs++;
+   *varying = tgsi_get_generic_gl_varying_index(aapoint_input->data.location, false);
+   state.input = aapoint_input;
+
+   nir_foreach_function(function, shader) {
+      if (function->impl) {
+         nir_lower_aapoint_impl(function->impl, &state);
+      }
+   }
+}
diff --git a/src/gallium/auxiliary/nir/nir_draw_helpers.h b/src/gallium/auxiliary/nir/nir_draw_helpers.h
new file mode 100644 (file)
index 0000000..08d163b
--- /dev/null
@@ -0,0 +1,40 @@
+/**************************************************************************
+ *
+ * Copyright 2019 Red Hat.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef NIR_DRAW_HELPERS_H
+#define NIR_DRAW_HELPERS_H
+struct nir_shader;
+void
+nir_lower_pstipple_fs(struct nir_shader *shader,
+                      unsigned *samplerUnitOut,
+                      unsigned fixedUnit,
+                      bool fs_pos_is_sysval);
+
+void
+nir_lower_aaline_fs(struct nir_shader *shader, int *varying);
+
+void
+nir_lower_aapoint_fs(struct nir_shader *shader, int *varying);
+#endif