WIP: glsl: Add pass to replace vector-scalar operations with vector-vector operations

4 files changed, 104 insertions, 0 deletions

diff --git a/src/compiler/Makefile.sources b/src/compiler/Makefile.sources
index f5193f6e2ec..7380170f1d1 100644
--- a/src/compiler/Makefile.sources
+++ b/src/compiler/Makefile.sources
@@ -83,6 +83,7 @@ LIBGLSL_FILES = \
 	glsl/loop_analysis.cpp \
 	glsl/loop_analysis.h \
 	glsl/loop_unroll.cpp \
+	glsl/lower_arithmetic_width.cpp \
 	glsl/lower_blend_equation_advanced.cpp \
 	glsl/lower_buffer_access.cpp \
 	glsl/lower_buffer_access.h \
diff --git a/src/compiler/glsl/ir_optimization.h b/src/compiler/glsl/ir_optimization.h
index 81049a479e8..bd0c643b903 100644
--- a/src/compiler/glsl/ir_optimization.h
+++ b/src/compiler/glsl/ir_optimization.h
@@ -180,4 +180,6 @@ ir_variable *compare_index_block(ir_builder::ir_factory &body,
 bool lower_64bit_integer_instructions(exec_list *instructions,
                                       unsigned what_to_lower);
 
+void lower_arithmetic_width(exec_list *instructions);
+
 #endif /* GLSL_IR_OPTIMIZATION_H */
diff --git a/src/compiler/glsl/lower_arithmetic_width.cpp b/src/compiler/glsl/lower_arithmetic_width.cpp
new file mode 100644
index 00000000000..5d2028d0ba0
--- /dev/null
+++ b/src/compiler/glsl/lower_arithmetic_width.cpp
@@ -0,0 +1,100 @@
+/*
+ * Copyright © 2017 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.
+ */
+
+/**
+ * \name lower_arithmetic_width.cpp
+ *
+ * GLSL IR allows mixing of scalars and either vectors or matrices for some
+ * arithmetic operations.  Vector architectures implement these operations by
+ * swizzling the scalar to a vector of the correct size.  In addition, SPIR-V
+ * does not allow mixed scalar and vector operations for anything except
+ * multiplication.
+ *
+ * This pass replaces mixed scalar-vector operations with vector-vector
+ * operations by applying a swizzle to the scalar operand to create a vector
+ * of matching size.
+ *
+ * If the reduce_multiply option is specified, multiplications with a vector
+ * operand that is effectively a scalar will be converted to a scalar (e.g.,
+ * foo.yyyy becomes foo.y).  If the operand is a scalar, the swizzle is
+ * removed (e.g., scalar_foo.xxx becomes scalar_foo).
+ */
+
+#include "ir.h"
+#include "ir_hierarchical_visitor.h"
+
+class arithmetic_width_visitor : public ir_hierarchical_visitor {
+public:
+   arithmetic_width_visitor()
+   {
+      /* empty */
+   }
+
+   ir_visitor_status visit_leave(ir_expression *);
+};
+
+ir_visitor_status
+arithmetic_width_visitor::visit_leave(ir_expression *ir)
+{
+   /* If there's only one operand, there's nothing to mix. */
+   if (ir->num_operands == 1)
+      return visit_continue;
+
+   /* All operations where GLSL IR allows mixing are operations that result in
+    * a vector.
+    */
+   if (!ir->type->is_vector())
+      return visit_continue;
+
+   if (ir->operation == ir_binop_mul &&
+       (ir->type->is_float() || ir->type->is_double())) {
+      return visit_continue;
+   }
+
+   const unsigned vector_size = ir->type->vector_elements;
+   const struct ir_swizzle_mask swiz_mask = { 0, 0, 0, 0, vector_size, true };
+   void *const mem_ctx = ralloc_parent(ir);
+
+   for (unsigned i = 0; i < ir->num_operands; ++i) {
+      ir_rvalue *const src = ir->operands[i];
+
+      assert(!src->type->is_vector() ||
+             vector_size == src->type->vector_elements);
+
+      if (src->type->is_scalar()) {
+         ir_swizzle *const swiz = new(mem_ctx) ir_swizzle(src, swiz_mask);
+
+         ir->operands[i] = swiz;
+      }
+   }
+
+   return visit_continue;
+}
+
+void
+lower_arithmetic_width(exec_list *instructions)
+{
+   arithmetic_width_visitor v;
+
+   v.run(instructions);
+}
diff --git a/src/compiler/glsl/meson.build b/src/compiler/glsl/meson.build
index 25300c77935..6fb3dcbec1d 100644
--- a/src/compiler/glsl/meson.build
+++ b/src/compiler/glsl/meson.build
@@ -124,6 +124,7 @@ files_libglsl = files(
   'loop_analysis.cpp',
   'loop_analysis.h',
   'loop_unroll.cpp',
+  'lower_arithmetic_width.cpp',
   'lower_blend_equation_advanced.cpp',
   'lower_buffer_access.cpp',
   'lower_buffer_access.h',