nir: add AMD_gcn_shader extended instructions

Signed-off-by: Daniel Schürmann <daniel.schuermann@campus.tu-berlin.de>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

diff --git a/src/compiler/nir/nir_lower_alu_to_scalar.c b/src/compiler/nir/nir_lower_alu_to_scalar.c
index 080d980f3479bc3c6dcf2166d885d938e6bde2b4..a0377dcb0be9b2c59feacadcb7a226116333ec70 100644 (file)
--- a/src/compiler/nir/nir_lower_alu_to_scalar.c
+++ b/src/compiler/nir/nir_lower_alu_to_scalar.c
@@ -96,6 +96,8 @@ lower_alu_instr_scalar(nir_alu_instr *instr, nir_builder *b)
    case nir_op_vec4:
    case nir_op_vec3:
    case nir_op_vec2:
+   case nir_op_cube_face_coord:
+   case nir_op_cube_face_index:
       /* We don't need to scalarize these ops, they're the ones generated to
        * group up outputs into a value that can be SSAed.
        */

diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_opcodes.py
index 97da4db28f16b789c33077a701bc24e970cebfbc..65d13200624066b56a986aa41b646016a7dcb286 100644 (file)
--- a/src/compiler/nir/nir_opcodes.py
+++ b/src/compiler/nir/nir_opcodes.py
@@ -352,6 +352,34 @@ for i in xrange(1, 5):
    for j in xrange(1, 5):
       unop_horiz("fnoise{0}_{1}".format(i, j), i, tfloat, j, tfloat, "0.0f")
 
+
+# AMD_gcn_shader extended instructions
+unop_horiz("cube_face_coord", 2, tfloat32, 3, tfloat32, """
+dst.x = dst.y = 0.0;
+float absX = fabs(src0.x);
+float absY = fabs(src0.y);
+float absZ = fabs(src0.z);
+if (src0.x >= 0 && absX >= absY && absX >= absZ) { dst.x = -src0.y; dst.y = -src0.z; }
+if (src0.x < 0 && absX >= absY && absX >= absZ) { dst.x = -src0.y; dst.y = src0.z; }
+if (src0.y >= 0 && absY >= absX && absY >= absZ) { dst.x = src0.z; dst.y = src0.x; }
+if (src0.y < 0 && absY >= absX && absY >= absZ) { dst.x = -src0.z; dst.y = src0.x; }
+if (src0.z >= 0 && absZ >= absX && absZ >= absY) { dst.x = -src0.y; dst.y = src0.x; }
+if (src0.z < 0 && absZ >= absX && absZ >= absY) { dst.x = -src0.y; dst.y = -src0.x; }
+""")
+
+unop_horiz("cube_face_index", 1, tfloat32, 3, tfloat32, """
+float absX = fabs(src0.x);
+float absY = fabs(src0.y);
+float absZ = fabs(src0.z);
+if (src0.x >= 0 && absX >= absY && absX >= absZ) dst.x = 0;
+if (src0.x < 0 && absX >= absY && absX >= absZ) dst.x = 1;
+if (src0.y >= 0 && absY >= absX && absY >= absZ) dst.x = 2;
+if (src0.y < 0 && absY >= absX && absY >= absZ) dst.x = 3;
+if (src0.z >= 0 && absZ >= absX && absZ >= absY) dst.x = 4;
+if (src0.z < 0 && absZ >= absX && absZ >= absY) dst.x = 5;
+""")
+
+
 def binop_convert(name, out_type, in_type, alg_props, const_expr):
    opcode(name, 0, out_type, [0, 0], [in_type, in_type], alg_props, const_expr)