nir: Add a lowering pass for non-uniform resource access

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

src/compiler/Makefile.sources

@@ -255,6 +255,7 @@ NIR_FILES = \
 	nir/nir_lower_io_to_temporaries.c \
 	nir/nir_lower_io_to_scalar.c \
 	nir/nir_lower_io_to_vector.c \
+	nir/nir_lower_non_uniform_access.c \
 	nir/nir_lower_packing.c \
 	nir/nir_lower_passthrough_edgeflags.c \
 	nir/nir_lower_patch_vertices.c \

src/compiler/nir/meson.build

@@ -136,6 +136,7 @@ files_libnir = files(
   'nir_lower_io_to_temporaries.c',
   'nir_lower_io_to_scalar.c',
   'nir_lower_io_to_vector.c',
+  'nir_lower_non_uniform_access.c',
   'nir_lower_packing.c',
   'nir_lower_passthrough_edgeflags.c',
   'nir_lower_patch_vertices.c',

src/compiler/nir/nir.h

@@ -1485,6 +1485,12 @@ typedef struct {
    /* gather offsets */
    int8_t tg4_offsets[4][2];
 
+   /* True if the texture index or handle is not dynamically uniform */
+   bool texture_non_uniform;
+
+   /* True if the sampler index or handle is not dynamically uniform */
+   bool sampler_non_uniform;
+
    /** The texture index
     *
     * If this texture instruction has a nir_tex_src_texture_offset source,
@@ -3259,6 +3265,16 @@ typedef struct nir_lower_tex_options {
 bool nir_lower_tex(nir_shader *shader,
                    const nir_lower_tex_options *options);
 
+enum nir_lower_non_uniform_access_type {
+   nir_lower_non_uniform_ubo_access     = (1 << 0),
+   nir_lower_non_uniform_ssbo_access    = (1 << 1),
+   nir_lower_non_uniform_texture_access = (1 << 2),
+   nir_lower_non_uniform_image_access   = (1 << 3),
+};
+
+bool nir_lower_non_uniform_access(nir_shader *shader,
+                                  enum nir_lower_non_uniform_access_type);
+
 bool nir_lower_idiv(nir_shader *shader);
 
 bool nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars);

src/compiler/nir/nir_lower_non_uniform_access.c

@@ -0,0 +1,265 @@
+/*
+ * Copyright © 2019 Intel Corporation
+ *
+ * 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 (including the next
+ * paragraph) 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.
+ */
+
+#include "nir.h"
+#include "nir_builder.h"
+
+static nir_ssa_def *
+read_first_invocation(nir_builder *b, nir_ssa_def *x)
+{
+   nir_intrinsic_instr *first =
+      nir_intrinsic_instr_create(b->shader,
+                                 nir_intrinsic_read_first_invocation);
+   first->num_components = x->num_components;
+   first->src[0] = nir_src_for_ssa(x);
+   nir_ssa_dest_init(&first->instr, &first->dest,
+                     x->num_components, x->bit_size, NULL);
+   return &first->dest.ssa;
+}
+
+static bool
+lower_non_uniform_tex_access(nir_builder *b, nir_tex_instr *tex)
+{
+   if (!tex->texture_non_uniform && !tex->sampler_non_uniform)
+      return false;
+
+   /* We can have at most one texture and one sampler handle */
+   nir_ssa_def *handles[2];
+   unsigned handle_count = 0;
+   for (unsigned i = 0; i < tex->num_srcs; i++) {
+      switch (tex->src[i].src_type) {
+      case nir_tex_src_texture_offset:
+      case nir_tex_src_texture_handle:
+         if (!tex->texture_non_uniform)
+            continue;
+         break;
+
+      case nir_tex_src_sampler_offset:
+      case nir_tex_src_sampler_handle:
+         if (!tex->sampler_non_uniform)
+            continue;
+         break;
+
+      default:
+         continue;
+      }
+
+      assert(tex->src[i].src.is_ssa);
+      assert(tex->src[i].src.ssa->num_components == 1);
+      assert(handle_count < 2);
+      handles[handle_count++] = tex->src[i].src.ssa;
+   }
+
+   if (handle_count == 0)
+      return false;
+
+   b->cursor = nir_instr_remove(&tex->instr);
+
+   nir_push_loop(b);
+
+   nir_ssa_def *all_equal_first = nir_imm_true(b);
+   for (unsigned i = 0; i < handle_count; i++) {
+      nir_ssa_def *equal_first =
+         nir_ieq(b, read_first_invocation(b, handles[i]), handles[i]);
+      all_equal_first = nir_iand(b, all_equal_first, equal_first);
+   }
+
+   nir_push_if(b, all_equal_first);
+
+   nir_builder_instr_insert(b, &tex->instr);
+   nir_jump(b, nir_jump_break);
+
+   return true;
+}
+
+static bool
+lower_non_uniform_access_intrin(nir_builder *b, nir_intrinsic_instr *intrin,
+                                unsigned handle_src)
+{
+   if (!(nir_intrinsic_access(intrin) & ACCESS_NON_UNIFORM))
+      return false;
+
+   /* If it's constant, it's automatically uniform; don't bother. */
+   if (nir_src_is_const(intrin->src[handle_src]))
+      return false;
+
+   b->cursor = nir_instr_remove(&intrin->instr);
+
+   nir_push_loop(b);
+
+   assert(intrin->src[handle_src].is_ssa);
+   assert(intrin->src[handle_src].ssa->num_components == 1);
+   nir_ssa_def *handle = intrin->src[handle_src].ssa;
+
+   nir_push_if(b, nir_ieq(b, read_first_invocation(b, handle), handle));
+
+   nir_builder_instr_insert(b, &intrin->instr);
+   nir_jump(b, nir_jump_break);
+
+   return true;
+}
+
+static bool
+nir_lower_non_uniform_access_impl(nir_function_impl *impl,
+                                  enum nir_lower_non_uniform_access_type types)
+{
+   bool progress = false;
+
+   nir_builder b;
+   nir_builder_init(&b, impl);
+
+   nir_foreach_block(block, impl) {
+      nir_foreach_instr(instr, block) {
+         switch (instr->type) {
+         case nir_instr_type_tex: {
+            nir_tex_instr *tex = nir_instr_as_tex(instr);
+            if ((types & nir_lower_non_uniform_texture_access) &&
+                lower_non_uniform_tex_access(&b, tex))
+               progress = true;
+            break;
+         }
+
+         case nir_instr_type_intrinsic: {
+            nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
+            switch (intrin->intrinsic) {
+            case nir_intrinsic_load_ubo:
+               if ((types & nir_lower_non_uniform_ubo_access) &&
+                   lower_non_uniform_access_intrin(&b, intrin, 0))
+                  progress = true;
+               break;
+
+            case nir_intrinsic_load_ssbo:
+            case nir_intrinsic_ssbo_atomic_add:
+            case nir_intrinsic_ssbo_atomic_imin:
+            case nir_intrinsic_ssbo_atomic_umin:
+            case nir_intrinsic_ssbo_atomic_imax:
+            case nir_intrinsic_ssbo_atomic_umax:
+            case nir_intrinsic_ssbo_atomic_and:
+            case nir_intrinsic_ssbo_atomic_or:
+            case nir_intrinsic_ssbo_atomic_xor:
+            case nir_intrinsic_ssbo_atomic_exchange:
+            case nir_intrinsic_ssbo_atomic_comp_swap:
+            case nir_intrinsic_ssbo_atomic_fadd:
+            case nir_intrinsic_ssbo_atomic_fmin:
+            case nir_intrinsic_ssbo_atomic_fmax:
+            case nir_intrinsic_ssbo_atomic_fcomp_swap:
+               if ((types & nir_lower_non_uniform_ssbo_access) &&
+                   lower_non_uniform_access_intrin(&b, intrin, 0))
+                  progress = true;
+               break;
+
+            case nir_intrinsic_store_ssbo:
+               /* SSBO Stores put the index in the second source */
+               if ((types & nir_lower_non_uniform_ssbo_access) &&
+                   lower_non_uniform_access_intrin(&b, intrin, 1))
+                  progress = true;
+               break;
+
+            case nir_intrinsic_image_load:
+            case nir_intrinsic_image_store:
+            case nir_intrinsic_image_atomic_add:
+            case nir_intrinsic_image_atomic_min:
+            case nir_intrinsic_image_atomic_max:
+            case nir_intrinsic_image_atomic_and:
+            case nir_intrinsic_image_atomic_or:
+            case nir_intrinsic_image_atomic_xor:
+            case nir_intrinsic_image_atomic_exchange:
+            case nir_intrinsic_image_atomic_comp_swap:
+            case nir_intrinsic_image_atomic_fadd:
+            case nir_intrinsic_image_size:
+            case nir_intrinsic_image_samples:
+            case nir_intrinsic_bindless_image_load:
+            case nir_intrinsic_bindless_image_store:
+            case nir_intrinsic_bindless_image_atomic_add:
+            case nir_intrinsic_bindless_image_atomic_min:
+            case nir_intrinsic_bindless_image_atomic_max:
+            case nir_intrinsic_bindless_image_atomic_and:
+            case nir_intrinsic_bindless_image_atomic_or:
+            case nir_intrinsic_bindless_image_atomic_xor:
+            case nir_intrinsic_bindless_image_atomic_exchange:
+            case nir_intrinsic_bindless_image_atomic_comp_swap:
+            case nir_intrinsic_bindless_image_atomic_fadd:
+            case nir_intrinsic_bindless_image_size:
+            case nir_intrinsic_bindless_image_samples:
+               if ((types & nir_lower_non_uniform_image_access) &&
+                   lower_non_uniform_access_intrin(&b, intrin, 0))
+                  progress = true;
+               break;
+
+            default:
+               /* Nothing to do */
+               break;
+            }
+            break;
+         }
+
+         default:
+            /* Nothing to do */
+            break;
+         }
+      }
+   }
+
+   if (progress)
+      nir_metadata_preserve(impl, nir_metadata_none);
+
+   return progress;
+}
+
+/**
+ * Lowers non-uniform resource access by using a loop
+ *
+ * This pass lowers non-uniform resource access by using subgroup operations
+ * and a loop.  Most hardware requires things like textures and UBO access
+ * operations to happen on a dynamically uniform (or at least subgroup
+ * uniform) resource.  This pass allows for non-uniform access by placing the
+ * texture instruction in a loop that looks something like this:
+ *
+ * loop {
+ *    bool tex_eq_first = readFirstInvocationARB(texture) == texture;
+ *    bool smp_eq_first = readFirstInvocationARB(sampler) == sampler;
+ *    if (tex_eq_first && smp_eq_first) {
+ *       res = texture(texture, sampler, ...);
+ *       break;
+ *    }
+ * }
+ *
+ * Fortunately, because the instruction is immediately followed by the only
+ * break in the loop, the block containing the instruction dominates the end
+ * of the loop.  Therefore, it's safe to move the instruction into the loop
+ * without fixing up SSA in any way.
+ */
+bool
+nir_lower_non_uniform_access(nir_shader *shader,
+                             enum nir_lower_non_uniform_access_type types)
+{
+   bool progress = false;
+
+   nir_foreach_function(function, shader) {
+      if (function->impl &&
+          nir_lower_non_uniform_access_impl(function->impl, types))
+         progress = true;
+   }
+
+   return progress;
+}

src/compiler/shader_enums.h

@@ -716,6 +716,9 @@ enum gl_access_qualifier
    ACCESS_VOLATILE      = (1 << 2),
    ACCESS_NON_READABLE  = (1 << 3),
    ACCESS_NON_WRITEABLE = (1 << 4),
+
+   /** The access may use a non-uniform buffer or image index */
+   ACCESS_NON_UNIFORM   = (1 << 5),
 };
 
 /**