[Mesa-dev] [PATCH 2/3] i965: Add a NIR-based cubemap normalizing pass

[Jason Ekstrand]

---
 src/mesa/drivers/dri/i965/Makefile.sources         |   1 +
 src/mesa/drivers/dri/i965/brw_nir.h                |   2 +
 .../drivers/dri/i965/brw_nir_cubemap_normalize.c   | 111 +++++++++++++++++++++
 3 files changed, 114 insertions(+)
 create mode 100644 src/mesa/drivers/dri/i965/brw_nir_cubemap_normalize.c

diff --git a/src/mesa/drivers/dri/i965/Makefile.sources b/src/mesa/drivers/dri/i965/Makefile.sources
index 498d5a7..8e723b0 100644
--- a/src/mesa/drivers/dri/i965/Makefile.sources
+++ b/src/mesa/drivers/dri/i965/Makefile.sources
@@ -78,6 +78,7 @@ i965_FILES = \
 	brw_multisample_state.h \
 	brw_nir.h \
 	brw_nir_analyze_boolean_resolves.c \
+	brw_nir_cubemap_normalize.c \
 	brw_object_purgeable.c \
 	brw_packed_float.c \
 	brw_performance_monitor.c \
diff --git a/src/mesa/drivers/dri/i965/brw_nir.h b/src/mesa/drivers/dri/i965/brw_nir.h
index 27782a3..5885a07 100644
--- a/src/mesa/drivers/dri/i965/brw_nir.h
+++ b/src/mesa/drivers/dri/i965/brw_nir.h
@@ -73,6 +73,8 @@ enum {
 
 void brw_nir_analyze_boolean_resolves(nir_shader *nir);
 
+void brw_nir_cubemap_normalize(nir_shader *nir);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/mesa/drivers/dri/i965/brw_nir_cubemap_normalize.c b/src/mesa/drivers/dri/i965/brw_nir_cubemap_normalize.c
new file mode 100644
index 0000000..6464f41
--- /dev/null
+++ b/src/mesa/drivers/dri/i965/brw_nir_cubemap_normalize.c
@@ -0,0 +1,111 @@
+/*
+ * Copyright © 2015 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.
+ *
+ * Authors:
+ *    Jason Ekstrand <jason at jlekstrand.net>
+ */
+
+#include "brw_nir.h"
+#include "glsl/nir/nir_builder.h"
+
+/**
+ * This file implements a NIR lowering pass to perform the normalization of
+ * the cubemap coordinates to have the largest magnitude component be -1.0
+ * or 1.0.  This is based on the old GLSL IR based pass by Eric.
+ */
+
+static nir_ssa_def *
+channel(nir_builder *b, nir_ssa_def *def, int c)
+{
+   return nir_swizzle(b, def, (unsigned[4]){c, c, c, c}, 1, false);
+}
+
+static bool
+cubemap_normalize_block(nir_block *block, void *void_state)
+{
+   nir_builder *b = void_state;
+
+   nir_foreach_instr(block, instr) {
+      if (instr->type != nir_instr_type_tex)
+         continue;
+
+      nir_tex_instr *tex = nir_instr_as_tex(instr);
+      if (tex->sampler_dim != GLSL_SAMPLER_DIM_CUBE)
+         continue;
+
+      nir_builder_insert_before_instr(b, &tex->instr);
+
+      for (unsigned i = 0; i < tex->num_srcs; i++) {
+         if (tex->src[i].src_type != nir_tex_src_coord)
+            continue;
+
+         nir_ssa_def *orig_coord =
+            nir_ssa_for_src(b, tex->src[i].src, nir_tex_instr_src_size(tex, i));
+         assert(orig_coord->num_components >= 3);
+
+         nir_ssa_def *abs0 = nir_fabs(b, channel(b, orig_coord, 0));
+         nir_ssa_def *abs1 = nir_fabs(b, channel(b, orig_coord, 1));
+         nir_ssa_def *abs2 = nir_fabs(b, channel(b, orig_coord, 2));
+
+         nir_ssa_def *norm1 = nir_fmax(b, abs0, nir_fmax(b, abs1, abs2));
+
+         nir_ssa_def *normalized = nir_fmul(b, orig_coord, nir_frcp(b, norm1));
+
+         /* Array indices don't have to be normalized, so make a new vector
+          * with the coordinate's array index untouched.
+          */
+         if (tex->coord_components == 4) {
+            normalized = nir_vec4(b,
+                                  channel(b, normalized, 0),
+                                  channel(b, normalized, 1),
+                                  channel(b, normalized, 2),
+                                  channel(b, orig_coord, 3));
+         }
+
+         nir_instr_rewrite_src(&tex->instr,
+                               &tex->src[i].src,
+                               nir_src_for_ssa(normalized));
+      }
+   }
+
+   return true;
+}
+
+static void
+cubemap_normalize_impl(nir_function_impl *impl)
+{
+   nir_builder b;
+   nir_builder_init(&b, impl);
+
+   nir_foreach_block(impl, cubemap_normalize_block, &b);
+
+   nir_metadata_preserve(impl, nir_metadata_block_index |
+                               nir_metadata_dominance);
+}
+
+void
+brw_nir_cubemap_normalize(nir_shader *shader)
+{
+   nir_foreach_overload(shader, overload)
+      if (overload->impl)
+         cubemap_normalize_impl(overload->impl);
+}
-- 
2.3.4