nir: Add more modulus and remainder opcodes

SPIR-V makes a distinction between "modulus" and "remainder" for both
floating-point and signed integer variants.  The difference is primarily
one of which source they take their sign from.  The "remainder" opcode for
integers is equivalent to the C/C++ "%" operation while the "modulus"
opcode is more mathematically correct (at least for an unsigned divisor).
This commit adds corresponding opcodes to NIR.

diff --git a/src/glsl/nir/nir_opcodes.py b/src/glsl/nir/nir_opcodes.py
index 9dbb341d91cd4625647c87dd74bf335de4b64374..d4f400d020da942d808ff1ad4dd3d50004a6c64d 100644 (file)
--- a/src/glsl/nir/nir_opcodes.py
+++ b/src/glsl/nir/nir_opcodes.py
@@ -369,9 +369,23 @@ binop_convert("uadd_carry", tuint, tuint, commutative, "src0 + src1 < src0")
 
 binop_convert("usub_borrow", tuint, tuint, "", "src0 < src1")
 
-binop("fmod", tfloat, "", "src0 - src1 * floorf(src0 / src1)")
 binop("umod", tuint, "", "src1 == 0 ? 0 : src0 % src1")
 
+# For signed integers, there are several different possible definitions of
+# "modulus" or "remainder".  We follow the conventions used by LLVM and
+# SPIR-V.  The irem opcode implements the standard C/C++ signed "%"
+# operation while the imod opcode implements the more mathematical
+# "modulus" operation.  For details on the difference, see
+#
+# http://mathforum.org/library/drmath/view/52343.html
+
+binop("irem", tint, "", "src1 == 0 ? 0 : src0 % src1")
+binop("imod", tint, "",
+      "src1 == 0 ? 0 : ((src0 % src1 == 0 || (src0 >= 0) == (src1 >= 0)) ?"
+      "                 src0 % src1 : src0 % src1 + src1)")
+binop("fmod", tfloat, "", "src0 - src1 * floorf(src0 / src1)")
+binop("frem", tfloat, "", "src0 - src1 * truncf(src0 / src1)")
+
 #
 # Comparisons
 #

diff --git a/src/glsl/nir/nir_opt_algebraic.py b/src/glsl/nir/nir_opt_algebraic.py
index db18fc9619ba1ab7e617fd9b77e4f455ba2d4888..416904669a4edbd7227e4604f2488a82cdff9b7f 100644 (file)
--- a/src/glsl/nir/nir_opt_algebraic.py
+++ b/src/glsl/nir/nir_opt_algebraic.py
@@ -226,6 +226,7 @@ optimizations = [
 
    # Misc. lowering
    (('fmod', a, b), ('fsub', a, ('fmul', b, ('ffloor', ('fdiv', a, b)))), 'options->lower_fmod'),
+   (('frem', a, b), ('fsub', a, ('fmul', b, ('ftrunc', ('fdiv', a, b)))), 'options->lower_fmod'),
    (('bitfield_insert', a, b, c, d), ('bfi', ('bfm', d, c), b, a), 'options->lower_bitfield_insert'),
    (('uadd_carry', a, b), ('b2i', ('ult', ('iadd', a, b), a)), 'options->lower_uadd_carry'),
    (('usub_borrow', a, b), ('b2i', ('ult', a, b)), 'options->lower_usub_borrow'),