nir: Add 1-bit Boolean opcodes

We also have to add support for 1-bit integers while we're here so we
get 1-bit variants of iand, ior, and inot.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-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 e406207c3cda6855d2af84c89704c4eec81ad074..9b175878c1549a042fd515db2db98b41bed5f32c 100644 (file)
--- a/src/compiler/nir/nir_lower_alu_to_scalar.c
+++ b/src/compiler/nir/nir_lower_alu_to_scalar.c
@@ -198,6 +198,10 @@ lower_alu_instr_scalar(nir_alu_instr *instr, nir_builder *b)
       return false;
 
       LOWER_REDUCTION(nir_op_fdot, nir_op_fmul, nir_op_fadd);
+      LOWER_REDUCTION(nir_op_ball_fequal, nir_op_feq, nir_op_iand);
+      LOWER_REDUCTION(nir_op_ball_iequal, nir_op_ieq, nir_op_iand);
+      LOWER_REDUCTION(nir_op_bany_fnequal, nir_op_fne, nir_op_ior);
+      LOWER_REDUCTION(nir_op_bany_inequal, nir_op_ine, nir_op_ior);
       LOWER_REDUCTION(nir_op_b32all_fequal, nir_op_feq32, nir_op_iand);
       LOWER_REDUCTION(nir_op_b32all_iequal, nir_op_ieq32, nir_op_iand);
       LOWER_REDUCTION(nir_op_b32any_fnequal, nir_op_fne32, nir_op_ior);

diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_opcodes.py
index c89bdeb436a5480495df79229f7071ff37ddd5dc..d32005846a625f6c333d2c1974a3c40528909f17 100644 (file)
--- a/src/compiler/nir/nir_opcodes.py
+++ b/src/compiler/nir/nir_opcodes.py
@@ -91,6 +91,7 @@ class Opcode(object):
 tfloat = "float"
 tint = "int"
 tbool = "bool"
+tbool1 = "bool1"
 tbool32 = "bool32"
 tuint = "uint"
 tuint16 = "uint16"
@@ -119,11 +120,11 @@ def type_sizes(type_):
     if type_has_size(type_):
         return [type_size(type_)]
     elif type_ == 'bool':
-        return [32]
+        return [1, 32]
     elif type_ == 'float':
         return [16, 32, 64]
     else:
-        return [8, 16, 32, 64]
+        return [1, 8, 16, 32, 64]
 
 def type_base_type(type_):
     m = _TYPE_SPLIT_RE.match(type_)
@@ -430,6 +431,9 @@ def binop_convert(name, out_type, in_type, alg_props, const_expr):
 def binop(name, ty, alg_props, const_expr):
    binop_convert(name, ty, ty, alg_props, const_expr)
 
+def binop_compare(name, ty, alg_props, const_expr):
+   binop_convert(name, tbool1, ty, alg_props, const_expr)
+
 def binop_compare32(name, ty, alg_props, const_expr):
    binop_convert(name, tbool32, ty, alg_props, const_expr)
 
@@ -550,6 +554,16 @@ binop("frem", tfloat, "", "src0 - src1 * truncf(src0 / src1)")
 
 # these integer-aware comparisons return a boolean (0 or ~0)
 
+binop_compare("flt", tfloat, "", "src0 < src1")
+binop_compare("fge", tfloat, "", "src0 >= src1")
+binop_compare("feq", tfloat, commutative, "src0 == src1")
+binop_compare("fne", tfloat, commutative, "src0 != src1")
+binop_compare("ilt", tint, "", "src0 < src1")
+binop_compare("ige", tint, "", "src0 >= src1")
+binop_compare("ieq", tint, commutative, "src0 == src1")
+binop_compare("ine", tint, commutative, "src0 != src1")
+binop_compare("ult", tuint, "", "src0 < src1")
+binop_compare("uge", tuint, "", "src0 >= src1")
 binop_compare32("flt32", tfloat, "", "src0 < src1")
 binop_compare32("fge32", tfloat, "", "src0 >= src1")
 binop_compare32("feq32", tfloat, commutative, "src0 == src1")
@@ -563,6 +577,15 @@ binop_compare32("uge32", tuint, "", "src0 >= src1")
 
 # integer-aware GLSL-style comparisons that compare floats and ints
 
+binop_reduce("ball_fequal",  1, tbool1, tfloat, "{src0} == {src1}",
+             "{src0} && {src1}", "{src}")
+binop_reduce("bany_fnequal", 1, tbool1, tfloat, "{src0} != {src1}",
+             "{src0} || {src1}", "{src}")
+binop_reduce("ball_iequal",  1, tbool1, tint, "{src0} == {src1}",
+             "{src0} && {src1}", "{src}")
+binop_reduce("bany_inequal", 1, tbool1, tint, "{src0} != {src1}",
+             "{src0} || {src1}", "{src}")
+
 binop_reduce("b32all_fequal",  1, tbool32, tfloat, "{src0} == {src1}",
              "{src0} && {src1}", "{src}")
 binop_reduce("b32any_fnequal", 1, tbool32, tfloat, "{src0} != {src1}",
@@ -756,6 +779,8 @@ triop("fmed3", tfloat, "fmaxf(fminf(fmaxf(src0, src1), src2), fminf(src0, src1))
 triop("imed3", tint, "MAX2(MIN2(MAX2(src0, src1), src2), MIN2(src0, src1))")
 triop("umed3", tuint, "MAX2(MIN2(MAX2(src0, src1), src2), MIN2(src0, src1))")
 
+opcode("bcsel", 0, tuint, [0, 0, 0],
+      [tbool1, tuint, tuint], "", "src0 ? src1 : src2")
 opcode("b32csel", 0, tuint, [0, 0, 0],
        [tbool32, tuint, tuint], "", "src0 ? src1 : src2")
 

diff --git a/src/compiler/nir/nir_search.c b/src/compiler/nir/nir_search.c
index 50f5464cef8feb40561d5255455d71395f7a0334..8578ca53c877ccc0d84d317315a807c6ae872ddb 100644 (file)
--- a/src/compiler/nir/nir_search.c
+++ b/src/compiler/nir/nir_search.c
@@ -110,7 +110,8 @@ nir_op_matches_search_op(nir_op nop, uint16_t sop)
 
 #define MATCH_BCONV_CASE(op) \
    case nir_search_op_##op: \
-      return nop == nir_op_##op##32;
+      return nop == nir_op_##op##1 || \
+             nop == nir_op_##op##32;
 
    switch (sop) {
    MATCH_FCONV_CASE(i2f)
@@ -160,6 +161,7 @@ nir_op_for_search_op(uint16_t sop, unsigned bit_size)
 #define RET_BCONV_CASE(op) \
    case nir_search_op_##op: \
       switch (bit_size) { \
+      case 1: return nir_op_##op##1; \
       case 32: return nir_op_##op##32; \
       default: unreachable("Invalid bit size"); \
       }