nir: Eliminate nir_op_f2b

Builds on the work of !15121.  This gets to delete even more code
because many drivers shared a lot of code for i2b and f2b.

No shader-db or fossil-db changes on any Intel platform.

v2: Rebase on 1a35acd8d9006c9.

v3: Update a comment in nir_opcodes_c.py. Suggested by Konstantin.

v4: Another rebase. Remove f2b stuff from Midgard.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20509>

diff --git a/src/amd/llvm/ac_nir_to_llvm.c b/src/amd/llvm/ac_nir_to_llvm.c
index 8fc7600..2b9d347 100644 (file)
--- a/src/amd/llvm/ac_nir_to_llvm.c
+++ b/src/amd/llvm/ac_nir_to_llvm.c
@@ -277,13 +277,6 @@ static LLVMValueRef emit_b2f(struct ac_llvm_context *ctx, LLVMValueRef src0, uns
    }
 }
 
-static LLVMValueRef emit_f2b(struct ac_llvm_context *ctx, LLVMValueRef src0)
-{
-   src0 = ac_to_float(ctx, src0);
-   LLVMValueRef zero = LLVMConstNull(LLVMTypeOf(src0));
-   return LLVMBuildFCmp(ctx->builder, LLVMRealUNE, src0, zero, "");
-}
-
 static LLVMValueRef emit_b2i(struct ac_llvm_context *ctx, LLVMValueRef src0, unsigned bitsize)
 {
    switch (bitsize) {
@@ -1097,9 +1090,6 @@ static bool visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr)
    case nir_op_b2f64:
       result = emit_b2f(&ctx->ac, src[0], instr->dest.dest.ssa.bit_size);
       break;
-   case nir_op_f2b1:
-      result = emit_f2b(&ctx->ac, src[0]);
-      break;
    case nir_op_b2i8:
    case nir_op_b2i16:
    case nir_op_b2i32:

diff --git a/src/asahi/compiler/agx_compile.c b/src/asahi/compiler/agx_compile.c
index d75e6ae..a07bc31 100644 (file)
--- a/src/asahi/compiler/agx_compile.c
+++ b/src/asahi/compiler/agx_compile.c
@@ -862,8 +862,6 @@ agx_emit_alu(agx_builder *b, nir_alu_instr *instr)
       else
          return agx_not_to(b, dst, s0);
 
-   case nir_op_f2b1:
-      return agx_fcmpsel_to(b, dst, s0, i0, i0, i1, AGX_FCOND_EQ);
    case nir_op_b2b1:
       return agx_icmpsel_to(b, dst, s0, i0, i0, i1, AGX_ICOND_UEQ);
 

diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index 6505fe5..725185f 100644 (file)
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -1210,11 +1210,6 @@ ntq_emit_comparison(struct v3d_compile *c,
                 vir_set_pf(c, vir_SUB_dest(c, nop, src0, src1), V3D_QPU_PF_PUSHC);
                 break;
 
-        case nir_op_f2b32:
-                vir_set_pf(c, vir_FMOV_dest(c, nop, src0), V3D_QPU_PF_PUSHZ);
-                cond_invert = true;
-                break;
-
         default:
                 return false;
         }
@@ -1589,7 +1584,6 @@ ntq_emit_alu(struct v3d_compile *c, nir_alu_instr *instr)
                 break;
         }
 
-        case nir_op_f2b32:
         case nir_op_feq32:
         case nir_op_fneu32:
         case nir_op_fge32:

diff --git a/src/compiler/glsl/float64.glsl b/src/compiler/glsl/float64.glsl
index e169053..5086c23 100644 (file)
--- a/src/compiler/glsl/float64.glsl
+++ b/src/compiler/glsl/float64.glsl
@@ -1174,12 +1174,6 @@ __int_to_fp64(int a)
    return __packFloat64(zSign, 0x412 - shiftCount, zFrac0, zFrac1);
 }
 
-bool
-__fp64_to_bool(uint64_t a)
-{
-   return !__feq64_nonnan(__fabs64(a), 0ul);
-}
-
 uint64_t
 __bool_to_fp64(bool a)
 {

diff --git a/src/compiler/nir/nir.c b/src/compiler/nir/nir.c
index ffdfc60..6fa1dfb 100644 (file)
--- a/src/compiler/nir/nir.c
+++ b/src/compiler/nir/nir.c
@@ -3109,10 +3109,6 @@ nir_alu_instr_is_comparison(const nir_alu_instr *instr)
    CASE_ALL_SIZES(nir_op_uge)
    CASE_ALL_SIZES(nir_op_ieq)
    CASE_ALL_SIZES(nir_op_ine)
-   case nir_op_f2b1:
-   case nir_op_f2b8:
-   case nir_op_f2b16:
-   case nir_op_f2b32:
    case nir_op_inot:
       return true;
    default:

diff --git a/src/compiler/nir/nir_algebraic.py b/src/compiler/nir/nir_algebraic.py
index bc4d88c..4bda78d 100644 (file)
--- a/src/compiler/nir/nir_algebraic.py
+++ b/src/compiler/nir/nir_algebraic.py
@@ -939,12 +939,12 @@ class TreeAutomaton(object):
             stripped = opcode.rstrip('0123456789')
             if stripped in conv_opcode_types:
                # Matches that use conversion opcodes with a specific type,
-               # like f2b1, are tricky.  Either we construct the automaton to
-               # match specific NIR opcodes like nir_op_f2b1, in which case we
+               # like f2i1, are tricky.  Either we construct the automaton to
+               # match specific NIR opcodes like nir_op_f2i1, in which case we
                # need to create separate items for each possible NIR opcode
-               # for patterns that have a generic opcode like f2b, or we
+               # for patterns that have a generic opcode like f2i, or we
                # construct it to match the search opcode, in which case we
-               # need to map f2b1 to f2b when constructing the automaton. Here
+               # need to map f2i1 to f2i when constructing the automaton. Here
                # we do the latter.
                opcode = stripped
             self.opcodes.add(opcode)

diff --git a/src/compiler/nir/nir_lower_bool_to_bitsize.c b/src/compiler/nir/nir_lower_bool_to_bitsize.c
index aa4f3de..47ae5d9 100644 (file)
--- a/src/compiler/nir/nir_lower_bool_to_bitsize.c
+++ b/src/compiler/nir/nir_lower_bool_to_bitsize.c
@@ -163,11 +163,6 @@ lower_alu_instr(nir_builder *b, nir_alu_instr *alu)
       /* Nothing to do here, we do not specialize these opcodes by bit-size */
       break;
 
-   case nir_op_f2b1:
-      opcode = bit_size == 8 ? nir_op_f2b8 :
-                               bit_size == 16 ? nir_op_f2b16 : nir_op_f2b32;
-      break;
-
    case nir_op_b2b1:
       /* Since the canonical bit size is the size of the src, it's a no-op */
       opcode = nir_op_mov;

diff --git a/src/compiler/nir/nir_lower_bool_to_float.c b/src/compiler/nir/nir_lower_bool_to_float.c
index 8ede52b..2d672f9 100644 (file)
--- a/src/compiler/nir/nir_lower_bool_to_float.c
+++ b/src/compiler/nir/nir_lower_bool_to_float.c
@@ -67,10 +67,6 @@ lower_alu_instr(nir_builder *b, nir_alu_instr *alu, bool has_fcsel_ne,
 
    case nir_op_b2f32: alu->op = nir_op_mov; break;
    case nir_op_b2i32: alu->op = nir_op_mov; break;
-   case nir_op_f2b1:
-      rep = nir_sne(b, nir_ssa_for_alu_src(b, alu, 0),
-                       nir_imm_float(b, 0));
-      break;
    case nir_op_b2b1: alu->op = nir_op_mov; break;
 
    case nir_op_flt: alu->op = nir_op_slt; break;

diff --git a/src/compiler/nir/nir_lower_bool_to_int32.c b/src/compiler/nir/nir_lower_bool_to_int32.c
index e9e21a6..779520e 100644 (file)
--- a/src/compiler/nir/nir_lower_bool_to_int32.c
+++ b/src/compiler/nir/nir_lower_bool_to_int32.c
@@ -66,8 +66,6 @@ lower_alu_instr(nir_alu_instr *alu)
       /* These we expect to have booleans but the opcode doesn't change */
       break;
 
-   case nir_op_f2b1: alu->op = nir_op_f2b32; break;
-
    case nir_op_b2b32:
    case nir_op_b2b1:
       /* We're mutating instructions in a dominance-preserving order so our

diff --git a/src/compiler/nir/nir_lower_double_ops.c b/src/compiler/nir/nir_lower_double_ops.c
index 91003e1..6615c77 100644 (file)
--- a/src/compiler/nir/nir_lower_double_ops.c
+++ b/src/compiler/nir/nir_lower_double_ops.c
@@ -504,12 +504,6 @@ lower_doubles_instr_to_soft(nir_builder *b, nir_alu_instr *instr,
       mangled_name = "__fp64_to_uint(u641;";
       return_type = glsl_uint_type();
       break;
-   case nir_op_f2b1:
-   case nir_op_f2b32:
-      name = "__fp64_to_bool";
-      mangled_name = "__fp64_to_bool(u641;";
-      return_type = glsl_bool_type();
-      break;
    case nir_op_b2f64:
       name = "__bool_to_fp64";
       mangled_name = "__bool_to_fp64(b1;";

diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_opcodes.py
index 2a3ea79..d8b80f6 100644 (file)
--- a/src/compiler/nir/nir_opcodes.py
+++ b/src/compiler/nir/nir_opcodes.py
@@ -233,7 +233,7 @@ for src_t in [tint, tuint, tfloat, tbool]:
    elif src_t == tuint:
       dst_types = [tfloat, tuint]
    elif src_t == tfloat:
-      dst_types = [tint, tuint, tfloat, tbool]
+      dst_types = [tint, tuint, tfloat]
 
    for dst_t in dst_types:
       for dst_bit_size in type_sizes(dst_t):

diff --git a/src/compiler/nir/nir_opcodes_c.py b/src/compiler/nir/nir_opcodes_c.py
index a51438a..2909a15 100644 (file)
--- a/src/compiler/nir/nir_opcodes_c.py
+++ b/src/compiler/nir/nir_opcodes_c.py
@@ -50,9 +50,10 @@ nir_type_conversion_op(nir_alu_type src, nir_alu_type dst, nir_rounding_mode rnd
       return nir_op_mov;
    }
 
-   /* i2b and u2b do not exist.  Use ine (via nir_type_conversion) instead */
-   assert((src_base != nir_type_int && src_base != nir_type_uint) ||
-          dst_base != nir_type_bool);
+   /* f2b, i2b, and u2b do not exist.  Use ine or fne (via nir_type_conversion)
+    * instead.
+    */
+   assert(src_base == dst_base || dst_base != nir_type_bool);
 
    switch (src_base) {
 %     for src_t in ['int', 'uint', 'float', 'bool']:
@@ -72,7 +73,7 @@ nir_type_conversion_op(nir_alu_type src, nir_alu_type dst, nir_rounding_mode rnd
 %                 else:
 <%                   dst_t = 'int' %>
 %                 endif
-%              elif src_t in ['int', 'uint'] and dst_t == 'bool':
+%              elif src_t != 'bool' and dst_t == 'bool':
 <%                   continue %>
 %              endif
                switch (dst_bit_size) {

diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index e56927e..cd83ded 100644 (file)
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -135,7 +135,6 @@ optimizations = [
    (('fabs', ('u2f', a)), ('u2f', a)),
    (('iabs', ('iabs', a)), ('iabs', a)),
    (('iabs', ('ineg', a)), ('iabs', a)),
-   (('f2b', ('fneg', a)), ('f2b', a)),
    (('~fadd', a, 0.0), a),
    # a+0.0 is 'a' unless 'a' is denormal or -0.0. If it's only used by a
    # floating point instruction, they should flush any input denormals and we
@@ -1452,7 +1451,6 @@ optimizations.extend([
    # Conversions
    (('f2i', ('ftrunc', a)), ('f2i', a)),
    (('f2u', ('ftrunc', a)), ('f2u', a)),
-   (('inot', ('f2b1', a)), ('feq', a, 0.0)),
 
    # Conversions from 16 bits to 32 bits and back can always be removed
    (('f2fmp', ('f2f32', 'a@16')), a),
@@ -2198,12 +2196,6 @@ for left, right in itertools.combinations_with_replacement(invert.keys(), 2):
    optimizations.append((('inot', ('iand(is_used_once)', (left, a, b), (right, c, d))),
                          ('ior', (invert[left], a, b), (invert[right], c, d))))
 
-# Optimize f2bN(b2f(x)) -> x
-for size in type_sizes('bool'):
-    aN = 'a@' + str(size)
-    f2bN = 'f2b' + str(size)
-    optimizations.append(((f2bN, ('b2f', aN)), a))
-
 # Optimize x2yN(b2x(x)) -> b2y
 for x, y in itertools.product(['f', 'u', 'i'], ['f', 'u', 'i']):
    if x != 'f' and y != 'f' and x != y:

diff --git a/src/compiler/nir/nir_search.c b/src/compiler/nir/nir_search.c
index 17f703a..e5b3c56 100644 (file)
--- a/src/compiler/nir/nir_search.c
+++ b/src/compiler/nir/nir_search.c
@@ -129,11 +129,6 @@ nir_op_matches_search_op(nir_op nop, uint16_t sop)
              nop == nir_op_##op##32 || \
              nop == nir_op_##op##64;
 
-#define MATCH_BCONV_CASE(op) \
-   case nir_search_op_##op: \
-      return nop == nir_op_##op##1 || \
-             nop == nir_op_##op##32;
-
    switch (sop) {
    MATCH_FCONV_CASE(i2f)
    MATCH_FCONV_CASE(u2f)
@@ -144,14 +139,12 @@ nir_op_matches_search_op(nir_op nop, uint16_t sop)
    MATCH_ICONV_CASE(i2i)
    MATCH_FCONV_CASE(b2f)
    MATCH_ICONV_CASE(b2i)
-   MATCH_BCONV_CASE(f2b)
    default:
       unreachable("Invalid nir_search_op");
    }
 
 #undef MATCH_FCONV_CASE
 #undef MATCH_ICONV_CASE
-#undef MATCH_BCONV_CASE
 }
 
 uint16_t
@@ -170,11 +163,6 @@ nir_search_op_for_nir_op(nir_op nop)
    case nir_op_##op##64: \
       return nir_search_op_##op;
 
-#define MATCH_BCONV_CASE(op) \
-   case nir_op_##op##1: \
-   case nir_op_##op##32: \
-      return nir_search_op_##op;
-
 
    switch (nop) {
    MATCH_FCONV_CASE(i2f)
@@ -186,14 +174,12 @@ nir_search_op_for_nir_op(nir_op nop)
    MATCH_ICONV_CASE(i2i)
    MATCH_FCONV_CASE(b2f)
    MATCH_ICONV_CASE(b2i)
-   MATCH_BCONV_CASE(f2b)
    default:
       return nop;
    }
 
 #undef MATCH_FCONV_CASE
 #undef MATCH_ICONV_CASE
-#undef MATCH_BCONV_CASE
 }
 
 static nir_op
@@ -221,14 +207,6 @@ nir_op_for_search_op(uint16_t sop, unsigned bit_size)
       default: unreachable("Invalid 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"); \
-      }
-
    switch (sop) {
    RET_FCONV_CASE(i2f)
    RET_FCONV_CASE(u2f)
@@ -239,14 +217,12 @@ nir_op_for_search_op(uint16_t sop, unsigned bit_size)
    RET_ICONV_CASE(i2i)
    RET_FCONV_CASE(b2f)
    RET_ICONV_CASE(b2i)
-   RET_BCONV_CASE(f2b)
    default:
       unreachable("Invalid nir_search_op");
    }
 
 #undef RET_FCONV_CASE
 #undef RET_ICONV_CASE
-#undef RET_BCONV_CASE
 }
 
 static bool
@@ -615,7 +591,6 @@ UNUSED static void dump_value(const nir_algebraic_table *table, const nir_search
       switch (expr->opcode) {
 #define CASE(n) \
       case nir_search_op_##n: fprintf(stderr, #n); break;
-      CASE(f2b)
       CASE(b2f)
       CASE(b2i)
       CASE(i2i)

diff --git a/src/compiler/nir/nir_search.h b/src/compiler/nir/nir_search.h
index 3b16778..953a245 100644 (file)
--- a/src/compiler/nir/nir_search.h
+++ b/src/compiler/nir/nir_search.h
@@ -123,7 +123,6 @@ enum nir_search_op {
    nir_search_op_i2i,
    nir_search_op_b2f,
    nir_search_op_b2i,
-   nir_search_op_f2b,
    nir_num_search_ops,
 };
 

diff --git a/src/freedreno/ir3/ir3_compiler_nir.c b/src/freedreno/ir3/ir3_compiler_nir.c
index d56a845..986762a 100644 (file)
--- a/src/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/freedreno/ir3/ir3_compiler_nir.c
@@ -469,12 +469,6 @@ emit_alu(struct ir3_context *ctx, nir_alu_instr *alu)
       dst[0] = create_cov(ctx, create_cov(ctx, src[0], 32, nir_op_f2f16_rtne),
                           16, nir_op_f2f32);
       break;
-   case nir_op_f2b1:
-      dst[0] = ir3_CMPS_F(
-         b, src[0], 0,
-         create_immed_typed(b, 0, type_float_size(bs[0])), 0);
-      dst[0]->cat2.condition = IR3_COND_NE;
-      break;
 
    case nir_op_b2b1:
       /* b2b1 will appear when translating from

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_nir.c b/src/gallium/auxiliary/gallivm/lp_bld_nir.c
index c2b82a7..3a127c2 100644 (file)
--- a/src/gallium/auxiliary/gallivm/lp_bld_nir.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_nir.c
@@ -245,23 +245,6 @@ assign_alu_dest(struct lp_build_nir_context *bld_base,
 }
 
 static LLVMValueRef
-flt_to_bool32(struct lp_build_nir_context *bld_base,
-              uint32_t src_bit_size,
-              LLVMValueRef val)
-{
-   LLVMBuilderRef builder = bld_base->base.gallivm->builder;
-   struct lp_build_context *flt_bld = get_flt_bld(bld_base, src_bit_size);
-   LLVMValueRef result =
-      lp_build_cmp(flt_bld, PIPE_FUNC_NOTEQUAL, val, flt_bld->zero);
-   if (src_bit_size == 64)
-      result = LLVMBuildTrunc(builder, result, bld_base->int_bld.vec_type, "");
-   if (src_bit_size == 16)
-      result = LLVMBuildSExt(builder, result, bld_base->int_bld.vec_type, "");
-   return result;
-}
-
-
-static LLVMValueRef
 fcmp32(struct lp_build_nir_context *bld_base,
        enum pipe_compare_func compare,
        uint32_t src_bit_size,
@@ -731,9 +714,6 @@ do_alu_action(struct lp_build_nir_context *bld_base,
    case nir_op_bitfield_reverse:
       result = lp_build_bitfield_reverse(get_int_bld(bld_base, false, src_bit_size[0]), src[0]);
       break;
-   case nir_op_f2b32:
-      result = flt_to_bool32(bld_base, src_bit_size[0], src[0]);
-      break;
    case nir_op_f2f16:
       if (src_bit_size[0] == 64)
          src[0] = LLVMBuildFPTrunc(builder, src[0],

diff --git a/src/gallium/auxiliary/nir/nir_to_tgsi.c b/src/gallium/auxiliary/nir/nir_to_tgsi.c
index f88f3ed..e2a4399 100644 (file)
--- a/src/gallium/auxiliary/nir/nir_to_tgsi.c
+++ b/src/gallium/auxiliary/nir/nir_to_tgsi.c
@@ -1561,13 +1561,6 @@ ntt_emit_alu(struct ntt_compile *c, nir_alu_instr *instr)
                   ntt_64bit_1f(c));
          break;
 
-      case nir_op_f2b32:
-         if (src_64)
-            ntt_DSNE(c, dst, src[0], ureg_imm1f(c->ureg, 0));
-         else
-            ntt_FSNE(c, dst, src[0], ureg_imm1f(c->ureg, 0));
-         break;
-
       case nir_op_b2i32:
          ntt_AND(c, dst, src[0], ureg_imm1u(c->ureg, 1));
          break;

diff --git a/src/gallium/drivers/etnaviv/etnaviv_compiler_nir_emit.c b/src/gallium/drivers/etnaviv/etnaviv_compiler_nir_emit.c
index f926057..54e8679 100644 (file)
--- a/src/gallium/drivers/etnaviv/etnaviv_compiler_nir_emit.c
+++ b/src/gallium/drivers/etnaviv/etnaviv_compiler_nir_emit.c
@@ -82,7 +82,6 @@ static const struct etna_op_info etna_ops[] = {
    UOP(f2u32, F2I, 0_X_X),
    UOP(b2f32, AND, 0_X_X), /* AND with fui(1.0f) */
    UOP(b2i32, AND, 0_X_X), /* AND with 1 */
-   OPC(f2b32, CMP, 0_X_X, NE), /* != 0.0 */
 
    /* arithmetic */
    IOP(iadd, ADD, 0_X_1),
@@ -163,9 +162,6 @@ etna_emit_alu(struct etna_compile *c, nir_op op, struct etna_inst_dst dst,
    case nir_op_b2i32:
       inst.src[2] = etna_immediate_int(1);
       break;
-   case nir_op_f2b32:
-      inst.src[1] = etna_immediate_float(0.0f);
-      break;
    case nir_op_ineg:
       inst.src[0] = etna_immediate_int(0);
       src[0].neg = 1;

diff --git a/src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp b/src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp
index 9400e69..1746a71 100644 (file)
--- a/src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp
+++ b/src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp
@@ -1552,8 +1552,6 @@ AluInstr::from_nir(nir_alu_instr *alu, Shader& shader)
    case nir_op_b32csel:
       return emit_alu_op3(*alu, op3_cnde_int, shader, {0, 2, 1});
 
-   case nir_op_f2b32:
-      return emit_alu_comb_with_zero(*alu, op2_setne_dx10, shader);
    case nir_op_fabs:
       return emit_alu_op1(*alu, op1_mov, shader, {1 << alu_src0_abs});
    case nir_op_fadd:

diff --git a/src/gallium/drivers/r600/sfn/sfn_nir_lower_64bit.cpp b/src/gallium/drivers/r600/sfn/sfn_nir_lower_64bit.cpp
index 09f294c..0657ce8 100644 (file)
--- a/src/gallium/drivers/r600/sfn/sfn_nir_lower_64bit.cpp
+++ b/src/gallium/drivers/r600/sfn/sfn_nir_lower_64bit.cpp
@@ -158,7 +158,6 @@ class LowerSplit64op : public NirLowerInstruction {
          switch (alu->op) {
          case nir_op_bcsel:
             return nir_dest_bit_size(alu->dest.dest) == 64;
-         case nir_op_f2b1:
          case nir_op_f2i32:
          case nir_op_f2u32:
          case nir_op_f2i64:
@@ -200,12 +199,6 @@ class LowerSplit64op : public NirLowerInstruction {
                          nir_unpack_64_2x32_split_y(b, nir_ssa_for_alu_src(b, alu, 2)));
             return nir_pack_64_2x32_split(b, lo, hi);
          }
-         case nir_op_f2b1: {
-            auto mask = nir_component_mask(nir_dest_num_components(alu->dest.dest));
-            return nir_fneu(b,
-                            nir_channels(b, nir_ssa_for_alu_src(b, alu, 0), mask),
-                            nir_imm_zero(b, nir_dest_num_components(alu->dest.dest), 64));
-         }
          case nir_op_f2i32: {
             auto src = nir_ssa_for_alu_src(b, alu, 0);
             auto gt0 = nir_flt(b, nir_imm_double(b, 0.0), src);

diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index c4461cf..52c4c86 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -1146,12 +1146,6 @@ ntq_emit_alu(struct vc4_compile *c, nir_alu_instr *instr)
         case nir_op_b2i32:
                 result = qir_AND(c, src[0], qir_uniform_ui(c, 1));
                 break;
-        case nir_op_f2b32:
-                qir_SF(c, src[0]);
-                result = qir_MOV(c, qir_SEL(c, QPU_COND_ZC,
-                                            qir_uniform_ui(c, ~0),
-                                            qir_uniform_ui(c, 0)));
-                break;
 
         case nir_op_iadd:
                 result = qir_ADD(c, src[0], src[1]);

diff --git a/src/gallium/drivers/zink/nir_to_spirv/nir_to_spirv.c b/src/gallium/drivers/zink/nir_to_spirv/nir_to_spirv.c
index 5a9d9af..365e1ca 100644 (file)
--- a/src/gallium/drivers/zink/nir_to_spirv/nir_to_spirv.c
+++ b/src/gallium/drivers/zink/nir_to_spirv/nir_to_spirv.c
@@ -2178,14 +2178,6 @@ emit_alu(struct ntv_context *ctx, nir_alu_instr *alu)
                           src[0]);
       break;
 
-   case nir_op_f2b1:
-      assert(nir_op_infos[alu->op].num_inputs == 1);
-      result = emit_binop(ctx, SpvOpFOrdNotEqual, dest_type, src[0],
-                          get_fvec_constant(ctx,
-                                            nir_src_bit_size(alu->src[0].src),
-                                            num_components, 0));
-      break;
-
 
 #define BINOP(nir_op, spirv_op) \
    case nir_op: \

diff --git a/src/intel/compiler/brw_fs_nir.cpp b/src/intel/compiler/brw_fs_nir.cpp
index 083e7fc..fe6023d 100644 (file)
--- a/src/intel/compiler/brw_fs_nir.cpp
+++ b/src/intel/compiler/brw_fs_nir.cpp
@@ -1571,34 +1571,6 @@ fs_visitor::nir_emit_alu(const fs_builder &bld, nir_alu_instr *instr,
       inst = bld.emit(SHADER_OPCODE_RSQ, result, op[0]);
       break;
 
-   case nir_op_f2b32: {
-      uint32_t bit_size = nir_src_bit_size(instr->src[0].src);
-      if (bit_size == 64) {
-         /* two-argument instructions can't take 64-bit immediates */
-         fs_reg zero = vgrf(glsl_type::double_type);
-         fs_reg tmp = vgrf(glsl_type::double_type);
-
-         bld.MOV(zero, setup_imm_df(bld, 0.0));
-
-         /* A SIMD16 execution needs to be split in two instructions, so use
-          * a vgrf instead of the flag register as dst so instruction splitting
-          * works
-          */
-         bld.CMP(tmp, op[0], zero, BRW_CONDITIONAL_NZ);
-         bld.MOV(result, subscript(tmp, BRW_REGISTER_TYPE_UD, 0));
-      } else {
-         fs_reg zero;
-         if (bit_size == 32) {
-            zero = brw_imm_f(0.0f);
-         } else {
-            assert(bit_size == 16);
-            zero = retype(brw_imm_w(0), BRW_REGISTER_TYPE_HF);
-         }
-         bld.CMP(result, op[0], zero, BRW_CONDITIONAL_NZ);
-      }
-      break;
-   }
-
    case nir_op_ftrunc:
       inst = bld.RNDZ(result, op[0]);
       if (devinfo->ver < 6) {

diff --git a/src/intel/compiler/brw_vec4_nir.cpp b/src/intel/compiler/brw_vec4_nir.cpp
index 2a3a3b9..e2f9230 100644 (file)
--- a/src/intel/compiler/brw_vec4_nir.cpp
+++ b/src/intel/compiler/brw_vec4_nir.cpp
@@ -1555,27 +1555,6 @@ vec4_visitor::nir_emit_alu(nir_alu_instr *instr)
       }
       break;
 
-   case nir_op_f2b32:
-      if (nir_src_bit_size(instr->src[0].src) == 64) {
-         /* We use a MOV with conditional_mod to check if the provided value is
-          * 0.0. We want this to flush denormalized numbers to zero, so we set a
-          * source modifier on the source operand to trigger this, as source
-          * modifiers don't affect the result of the testing against 0.0.
-          */
-         src_reg value = op[0];
-         value.abs = true;
-         vec4_instruction *inst = emit(MOV(dst_null_df(), value));
-         inst->conditional_mod = BRW_CONDITIONAL_NZ;
-
-         src_reg one = src_reg(this, glsl_type::ivec4_type);
-         emit(MOV(dst_reg(one), brw_imm_d(~0)));
-         inst = emit(BRW_OPCODE_SEL, dst, one, brw_imm_d(0));
-         inst->predicate = BRW_PREDICATE_NORMAL;
-      } else {
-         emit(CMP(dst, op[0], brw_imm_f(0.0f), BRW_CONDITIONAL_NZ));
-      }
-      break;
-
    case nir_op_unpack_half_2x16_split_x:
    case nir_op_unpack_half_2x16_split_y:
    case nir_op_pack_half_2x16_split:

diff --git a/src/microsoft/compiler/dxil_nir_algebraic.py b/src/microsoft/compiler/dxil_nir_algebraic.py
index c5f216c..9fd9ca5 100644 (file)
--- a/src/microsoft/compiler/dxil_nir_algebraic.py
+++ b/src/microsoft/compiler/dxil_nir_algebraic.py
@@ -94,7 +94,6 @@ remove_unsupported_casts(no_16bit_conv, 16, 0xffff, 65535.0, -32768.0, 32767.0)
 lower_x2b = [
   (('b2b32', 'a'), ('b2i32', 'a')),
   (('b2b1', 'a'), ('ine', ('b2i32', a), 0)),
-  (('f2b1', 'a'), ('fneu', a, 0)),
 ]
 
 no_16bit_conv += [

diff --git a/src/microsoft/compiler/nir_to_dxil.c b/src/microsoft/compiler/nir_to_dxil.c
index 695db9d..d690df4 100644 (file)
--- a/src/microsoft/compiler/nir_to_dxil.c
+++ b/src/microsoft/compiler/nir_to_dxil.c
@@ -2398,15 +2398,6 @@ emit_b2f64(struct ntd_context *ctx, nir_alu_instr *alu, const struct dxil_value
 }
 
 static bool
-emit_f2b32(struct ntd_context *ctx, nir_alu_instr *alu, const struct dxil_value *val)
-{
-   assert(val);
-
-   const struct dxil_value *zero = dxil_module_get_float_const(&ctx->mod, 0.0f);
-   return emit_cmp(ctx, alu, DXIL_FCMP_UNE, val, zero);
-}
-
-static bool
 emit_f16tof32(struct ntd_context *ctx, nir_alu_instr *alu, const struct dxil_value *val, bool shift)
 {
    if (shift) {
@@ -2716,7 +2707,6 @@ emit_alu(struct ntd_context *ctx, nir_alu_instr *alu)
    case nir_op_u2u64:
       return emit_cast(ctx, alu, src[0]);
 
-   case nir_op_f2b32: return emit_f2b32(ctx, alu, src[0]);
    case nir_op_b2f16: return emit_b2f16(ctx, alu, src[0]);
    case nir_op_b2f32: return emit_b2f32(ctx, alu, src[0]);
    case nir_op_b2f64: return emit_b2f64(ctx, alu, src[0]);

diff --git a/src/nouveau/codegen/nv50_ir_from_nir.cpp b/src/nouveau/codegen/nv50_ir_from_nir.cpp
index 7d18610..6970dcd 100644 (file)
--- a/src/nouveau/codegen/nv50_ir_from_nir.cpp
+++ b/src/nouveau/codegen/nv50_ir_from_nir.cpp
@@ -2921,12 +2921,6 @@ Converter::visit(nir_alu_instr *insn)
       mkOp2(OP_MERGE, TYPE_U64, newDefs[0], loadImm(NULL, 0), tmp);
       break;
    }
-   case nir_op_f2b32: {
-      DEFAULT_CHECKS;
-      LValues &newDefs = convert(&insn->dest);
-      mkCmp(OP_SET, CC_NEU, TYPE_U32, newDefs[0], sTypes[0], getSrc(&insn->src[0]), zero);
-      break;
-   }
    case nir_op_b2i8:
    case nir_op_b2i16:
    case nir_op_b2i32: {

diff --git a/src/panfrost/compiler/bifrost_compile.c b/src/panfrost/compiler/bifrost_compile.c
index ebcd30a..801803e 100644 (file)
--- a/src/panfrost/compiler/bifrost_compile.c
+++ b/src/panfrost/compiler/bifrost_compile.c
@@ -2880,14 +2880,6 @@ bi_emit_alu(bi_builder *b, nir_alu_instr *instr)
                 BI_MUX_INT_ZERO);
       break;
 
-   case nir_op_f2b16:
-      bi_mux_v2i16_to(b, dst, bi_imm_u16(0), bi_imm_u16(~0), s0,
-                      BI_MUX_FP_ZERO);
-      break;
-   case nir_op_f2b32:
-      bi_mux_i32_to(b, dst, bi_imm_u32(0), bi_imm_u32(~0), s0, BI_MUX_FP_ZERO);
-      break;
-
    case nir_op_ieq8:
    case nir_op_ine8:
    case nir_op_ilt8:

diff --git a/src/panfrost/midgard/midgard_compile.c b/src/panfrost/midgard/midgard_compile.c
index 4eca974..33855ee 100644 (file)
--- a/src/panfrost/midgard/midgard_compile.c
+++ b/src/panfrost/midgard/midgard_compile.c
@@ -836,11 +836,6 @@ emit_alu(compiler_context *ctx, nir_alu_instr *instr)
       ALU_CASE_CMP(b2f16, iand);
       ALU_CASE_CMP(b2i32, iand);
 
-      /* Likewise, we don't have a dedicated f2b32 instruction, but
-       * we can do a "not equal to 0.0" test. */
-
-      ALU_CASE_CMP(f2b32, fne);
-
       ALU_CASE(frcp, frcp);
       ALU_CASE(frsq, frsqrt);
       ALU_CASE(fsqrt, fsqrt);