[Mesa-dev] [PATCH 5/6] nir: Delete bany, ball, fany, fall.
Matt Turner
mattst88 at gmail.com
Mon Nov 30 15:32:10 PST 2015
As in the previous patches, these can be implemented as
any(v) -> any_nequal(v, false)
all(v) -> all_equal(v, true)
and their removal simplifies the code in the next patch.
---
src/gallium/auxiliary/nir/tgsi_to_nir.c | 4 +++-
src/glsl/nir/nir_lower_alu_to_scalar.c | 4 ----
src/glsl/nir/nir_opcodes.py | 7 -------
src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 6 ------
.../drivers/dri/i965/brw_nir_analyze_boolean_resolves.c | 3 ---
src/mesa/drivers/dri/i965/brw_vec4_nir.cpp | 14 --------------
src/mesa/program/prog_to_nir.c | 4 ++--
7 files changed, 5 insertions(+), 37 deletions(-)
diff --git a/src/gallium/auxiliary/nir/tgsi_to_nir.c b/src/gallium/auxiliary/nir/tgsi_to_nir.c
index 86c2ffa..87d702d 100644
--- a/src/gallium/auxiliary/nir/tgsi_to_nir.c
+++ b/src/gallium/auxiliary/nir/tgsi_to_nir.c
@@ -1062,7 +1062,9 @@ ttn_kill(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
static void
ttn_kill_if(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
{
- nir_ssa_def *cmp = nir_bany4(b, nir_flt(b, src[0], nir_imm_float(b, 0.0)));
+ nir_ssa_def *cmp = nir_bany_inequal4(b, nir_flt(b, src[0],
+ nir_imm_float(b, 0.0)),
+ nir_imm_int(b, 0));
nir_intrinsic_instr *discard =
nir_intrinsic_instr_create(b->shader, nir_intrinsic_discard_if);
discard->src[0] = nir_src_for_ssa(cmp);
diff --git a/src/glsl/nir/nir_lower_alu_to_scalar.c b/src/glsl/nir/nir_lower_alu_to_scalar.c
index 9313fc0..d267ca3 100644
--- a/src/glsl/nir/nir_lower_alu_to_scalar.c
+++ b/src/glsl/nir/nir_lower_alu_to_scalar.c
@@ -137,10 +137,6 @@ lower_alu_instr_scalar(nir_alu_instr *instr, nir_builder *b)
LOWER_REDUCTION(nir_op_bany_inequal, nir_op_ine, nir_op_ior);
LOWER_REDUCTION(nir_op_fall_equal, nir_op_seq, nir_op_fand);
LOWER_REDUCTION(nir_op_fany_nequal, nir_op_sne, nir_op_for);
- LOWER_REDUCTION(nir_op_ball, nir_op_imov, nir_op_iand);
- LOWER_REDUCTION(nir_op_bany, nir_op_imov, nir_op_ior);
- LOWER_REDUCTION(nir_op_fall, nir_op_fmov, nir_op_fand);
- LOWER_REDUCTION(nir_op_fany, nir_op_fmov, nir_op_for);
default:
break;
diff --git a/src/glsl/nir/nir_opcodes.py b/src/glsl/nir/nir_opcodes.py
index 37d3dfc..1cd01a4 100644
--- a/src/glsl/nir/nir_opcodes.py
+++ b/src/glsl/nir/nir_opcodes.py
@@ -167,13 +167,6 @@ unop_convert("i2b", tint, tbool, "src0 != 0")
unop_convert("b2i", tbool, tint, "src0 ? 1 : 0") # Boolean-to-int conversion
unop_convert("u2f", tuint, tfloat, "src0") # Unsigned-to-float conversion.
-unop_reduce("bany", 1, tbool, tbool, "{src}", "{src0} || {src1}", "{src}")
-unop_reduce("ball", 1, tbool, tbool, "{src}", "{src0} && {src1}", "{src}")
-unop_reduce("fany", 1, tfloat, tfloat, "{src} != 0.0f", "{src0} || {src1}",
- "{src} ? 1.0f : 0.0f")
-unop_reduce("fall", 1, tfloat, tfloat, "{src} != 0.0f", "{src0} && {src1}",
- "{src} ? 1.0f : 0.0f")
-
# Unary floating-point rounding operations.
diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index 9b50e4e..f5b203d 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
@@ -841,12 +841,6 @@ fs_visitor::nir_emit_alu(const fs_builder &bld, nir_alu_instr *instr)
case nir_op_fdot2:
case nir_op_fdot3:
case nir_op_fdot4:
- case nir_op_bany2:
- case nir_op_bany3:
- case nir_op_bany4:
- case nir_op_ball2:
- case nir_op_ball3:
- case nir_op_ball4:
case nir_op_ball_fequal2:
case nir_op_ball_iequal2:
case nir_op_ball_fequal3:
diff --git a/src/mesa/drivers/dri/i965/brw_nir_analyze_boolean_resolves.c b/src/mesa/drivers/dri/i965/brw_nir_analyze_boolean_resolves.c
index c995d2b..f4d23d8 100644
--- a/src/mesa/drivers/dri/i965/brw_nir_analyze_boolean_resolves.c
+++ b/src/mesa/drivers/dri/i965/brw_nir_analyze_boolean_resolves.c
@@ -109,9 +109,6 @@ analyze_boolean_resolves_block(nir_block *block, void *void_state)
uint8_t resolve_status;
nir_alu_instr *alu = nir_instr_as_alu(instr);
switch (alu->op) {
- case nir_op_bany2:
- case nir_op_bany3:
- case nir_op_bany4:
case nir_op_ball_fequal2:
case nir_op_ball_iequal2:
case nir_op_ball_fequal3:
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
index 4aed60e..6666457 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
@@ -1459,20 +1459,6 @@ vec4_visitor::nir_emit_alu(nir_alu_instr *instr)
inst->saturate = instr->dest.saturate;
break;
- case nir_op_bany2:
- case nir_op_bany3:
- case nir_op_bany4: {
- unsigned swiz =
- brw_swizzle_for_size(nir_op_infos[instr->op].input_sizes[0]);
-
- emit(CMP(dst_null_d(), swizzle(op[0], swiz), brw_imm_d(0),
- BRW_CONDITIONAL_NZ));
- emit(MOV(dst, brw_imm_d(0)));
- inst = emit(MOV(dst, brw_imm_d(~0)));
- inst->predicate = BRW_PREDICATE_ALIGN16_ANY4H;
- break;
- }
-
case nir_op_fabs:
case nir_op_iabs:
case nir_op_fneg:
diff --git a/src/mesa/program/prog_to_nir.c b/src/mesa/program/prog_to_nir.c
index 539e3c0..975bd1a 100644
--- a/src/mesa/program/prog_to_nir.c
+++ b/src/mesa/program/prog_to_nir.c
@@ -554,8 +554,8 @@ static void
ptn_kil(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
{
nir_ssa_def *cmp = b->shader->options->native_integers ?
- nir_bany4(b, nir_flt(b, src[0], nir_imm_float(b, 0.0))) :
- nir_fany4(b, nir_slt(b, src[0], nir_imm_float(b, 0.0)));
+ nir_bany_fnequal4(b, nir_flt(b, src[0], nir_imm_float(b, 0.0)), nir_imm_int(b, 0)) :
+ nir_fany_nequal4(b, nir_slt(b, src[0], nir_imm_float(b, 0.0)), nir_imm_float(b, 0.0));
nir_intrinsic_instr *discard =
nir_intrinsic_instr_create(b->shader, nir_intrinsic_discard_if);
--
2.4.9
More information about the mesa-dev
mailing list