[Mesa-dev] [PATCH 4/5] nir: Move vc4's alpha test lowering to core NIR.
Eric Anholt
eric at anholt.net
Tue Sep 19 18:06:11 UTC 2017
I've been doing this inside of vc4, but vc5 wants it as well and it may be
useful for other drivers (Intel has a related path for pre-gen6 with MRT,
and freedreno had a TGSI path for it at one point).
This required defining a common enum for the standard comparison
functions, but other lowering passes are likely to also want that enum.
---
src/compiler/Makefile.sources | 1 +
src/compiler/nir/nir.h | 2 +
src/compiler/nir/nir_builder.h | 25 ++++++
src/compiler/nir/nir_lower_alpha_test.c | 111 ++++++++++++++++++++++++++
src/compiler/shader_enums.h | 17 ++++
src/gallium/drivers/vc4/vc4_nir_lower_blend.c | 50 ------------
src/gallium/drivers/vc4/vc4_program.c | 15 +++-
src/gallium/drivers/vc4/vc4_qir.h | 1 -
8 files changed, 167 insertions(+), 55 deletions(-)
create mode 100644 src/compiler/nir/nir_lower_alpha_test.c
diff --git a/src/compiler/Makefile.sources b/src/compiler/Makefile.sources
index 0153df2d812c..c4c6ee38bb0b 100644
--- a/src/compiler/Makefile.sources
+++ b/src/compiler/Makefile.sources
@@ -207,6 +207,7 @@ NIR_FILES = \
nir/nir_loop_analyze.c \
nir/nir_loop_analyze.h \
nir/nir_lower_64bit_packing.c \
+ nir/nir_lower_alpha_test.c \
nir/nir_lower_alu_to_scalar.c \
nir/nir_lower_atomics.c \
nir/nir_lower_atomics_to_ssbo.c \
diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index 8330e6d7ce7e..0def0b586187 100644
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -2434,6 +2434,8 @@ bool nir_lower_constant_initializers(nir_shader *shader,
bool nir_move_vec_src_uses_to_dest(nir_shader *shader);
bool nir_lower_vec_to_movs(nir_shader *shader);
+void nir_lower_alpha_test(nir_shader *shader, enum compare_func func,
+ bool alpha_to_one);
bool nir_lower_alu_to_scalar(nir_shader *shader);
bool nir_lower_load_const_to_scalar(nir_shader *shader);
bool nir_lower_read_invocation_to_scalar(nir_shader *shader);
diff --git a/src/compiler/nir/nir_builder.h b/src/compiler/nir/nir_builder.h
index 7c65886356d5..4bd5628ff7d4 100644
--- a/src/compiler/nir/nir_builder.h
+++ b/src/compiler/nir/nir_builder.h
@@ -643,4 +643,29 @@ nir_jump(nir_builder *build, nir_jump_type jump_type)
nir_builder_instr_insert(build, &jump->instr);
}
+static inline nir_ssa_def *
+nir_compare_func(nir_builder *b, enum compare_func func,
+ nir_ssa_def *src0, nir_ssa_def *src1)
+{
+ switch (func) {
+ case COMPARE_FUNC_NEVER:
+ return nir_imm_int(b, 0);
+ case COMPARE_FUNC_ALWAYS:
+ return nir_imm_int(b, ~0);
+ case COMPARE_FUNC_EQUAL:
+ return nir_feq(b, src0, src1);
+ case COMPARE_FUNC_NOTEQUAL:
+ return nir_fne(b, src0, src1);
+ case COMPARE_FUNC_GREATER:
+ return nir_flt(b, src1, src0);
+ case COMPARE_FUNC_GEQUAL:
+ return nir_fge(b, src0, src1);
+ case COMPARE_FUNC_LESS:
+ return nir_flt(b, src0, src1);
+ case COMPARE_FUNC_LEQUAL:
+ return nir_fge(b, src1, src0);
+ }
+ unreachable("bad compare func");
+}
+
#endif /* NIR_BUILDER_H */
diff --git a/src/compiler/nir/nir_lower_alpha_test.c b/src/compiler/nir/nir_lower_alpha_test.c
new file mode 100644
index 000000000000..bd433b8ec66c
--- /dev/null
+++ b/src/compiler/nir/nir_lower_alpha_test.c
@@ -0,0 +1,111 @@
+/*
+ * Copyright © 2017 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+/**
+ * @file
+ *
+ * Implements GL alpha testing by comparing the output color's alpha to the
+ * alpha_ref intrinsic and emitting a discard based on it.
+ *
+ * The alpha_to_one value overrides the source alpha to 1.0 to implement
+ * GL_SAMPLE_ALPHA_TO_ONE, which applies before the alpha test (and would be
+ * rather silly to use with alpha test, but the spec permits).
+ */
+
+#include "nir/nir.h"
+#include "nir/nir_builder.h"
+
+void
+nir_lower_alpha_test(nir_shader *shader, enum compare_func func,
+ bool alpha_to_one)
+{
+ assert(shader->stage == MESA_SHADER_FRAGMENT);
+
+ nir_foreach_function(function, shader) {
+ nir_function_impl *impl = function->impl;
+ nir_builder b;
+ nir_builder_init(&b, impl);
+ b.cursor = nir_before_cf_list(&impl->body);
+
+ nir_foreach_block(block, impl) {
+ nir_foreach_instr_safe(instr, block) {
+ if (instr->type == nir_instr_type_intrinsic) {
+ nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
+
+ nir_variable *out = NULL;
+
+ switch (intr->intrinsic) {
+ case nir_intrinsic_store_var:
+ out = intr->variables[0]->var;
+ break;
+ case nir_intrinsic_store_output:
+ /* already had i/o lowered.. lookup the matching output var: */
+ nir_foreach_variable(var, &shader->outputs) {
+ int drvloc = var->data.driver_location;
+ if (nir_intrinsic_base(intr) == drvloc) {
+ out = var;
+ break;
+ }
+ }
+ assume(out);
+ break;
+ default:
+ continue;
+ }
+
+ if (out->data.mode != nir_var_shader_out)
+ continue;
+
+ if (out->data.location != FRAG_RESULT_COLOR &&
+ out->data.location != FRAG_RESULT_DATA0)
+ continue;
+
+ b.cursor = nir_before_instr(&intr->instr);
+
+ nir_ssa_def *alpha;
+ if (alpha_to_one) {
+ alpha = nir_imm_float(&b, 1.0);
+ } else {
+ alpha = nir_channel(&b, nir_ssa_for_src(&b, intr->src[0], 4),
+ 3);
+ }
+
+ nir_ssa_def *condition =
+ nir_compare_func(&b, func,
+ alpha, nir_load_alpha_ref_float(&b));
+
+ nir_intrinsic_instr *discard =
+ nir_intrinsic_instr_create(b.shader,
+ nir_intrinsic_discard_if);
+ discard->num_components = 1;
+ discard->src[0] = nir_src_for_ssa(nir_inot(&b, condition));
+ nir_builder_instr_insert(&b, &discard->instr);
+ shader->info.fs.uses_discard = true;
+ }
+ }
+ }
+
+ nir_metadata_preserve(impl, nir_metadata_block_index |
+ nir_metadata_dominance);
+ }
+}
diff --git a/src/compiler/shader_enums.h b/src/compiler/shader_enums.h
index 2f20e68c5d63..9d229d4199eb 100644
--- a/src/compiler/shader_enums.h
+++ b/src/compiler/shader_enums.h
@@ -670,6 +670,23 @@ enum gl_tess_spacing
TESS_SPACING_FRACTIONAL_EVEN,
};
+/**
+ * A compare function enum for use in compiler lowering passes. This is in
+ * the same order as GL's compare functions (shifted down by GL_NEVER), and is
+ * exactly the same as gallium's PIPE_FUNC_*.
+ */
+enum compare_func
+{
+ COMPARE_FUNC_NEVER,
+ COMPARE_FUNC_LESS,
+ COMPARE_FUNC_EQUAL,
+ COMPARE_FUNC_LEQUAL,
+ COMPARE_FUNC_GREATER,
+ COMPARE_FUNC_NOTEQUAL,
+ COMPARE_FUNC_GEQUAL,
+ COMPARE_FUNC_ALWAYS,
+};
+
#ifdef __cplusplus
} /* extern "C" */
#endif
diff --git a/src/gallium/drivers/vc4/vc4_nir_lower_blend.c b/src/gallium/drivers/vc4/vc4_nir_lower_blend.c
index a28ebb5bb7c7..60eccb4fc009 100644
--- a/src/gallium/drivers/vc4/vc4_nir_lower_blend.c
+++ b/src/gallium/drivers/vc4/vc4_nir_lower_blend.c
@@ -449,54 +449,6 @@ vc4_logicop(nir_builder *b, int logicop_func,
}
}
-static nir_ssa_def *
-vc4_nir_pipe_compare_func(nir_builder *b, int func,
- nir_ssa_def *src0, nir_ssa_def *src1)
-{
- switch (func) {
- default:
- fprintf(stderr, "Unknown compare func %d\n", func);
- /* FALLTHROUGH */
- case PIPE_FUNC_NEVER:
- return nir_imm_int(b, 0);
- case PIPE_FUNC_ALWAYS:
- return nir_imm_int(b, ~0);
- case PIPE_FUNC_EQUAL:
- return nir_feq(b, src0, src1);
- case PIPE_FUNC_NOTEQUAL:
- return nir_fne(b, src0, src1);
- case PIPE_FUNC_GREATER:
- return nir_flt(b, src1, src0);
- case PIPE_FUNC_GEQUAL:
- return nir_fge(b, src0, src1);
- case PIPE_FUNC_LESS:
- return nir_flt(b, src0, src1);
- case PIPE_FUNC_LEQUAL:
- return nir_fge(b, src1, src0);
- }
-}
-
-static void
-vc4_nir_emit_alpha_test_discard(struct vc4_compile *c, nir_builder *b,
- nir_ssa_def *alpha)
-{
- if (!c->fs_key->alpha_test)
- return;
-
- nir_ssa_def *condition =
- vc4_nir_pipe_compare_func(b, c->fs_key->alpha_test_func,
- alpha,
- nir_load_alpha_ref_float(b));
-
- nir_intrinsic_instr *discard =
- nir_intrinsic_instr_create(b->shader,
- nir_intrinsic_discard_if);
- discard->num_components = 1;
- discard->src[0] = nir_src_for_ssa(nir_inot(b, condition));
- nir_builder_instr_insert(b, &discard->instr);
- c->s->info.fs.uses_discard = true;
-}
-
static nir_ssa_def *
vc4_nir_swizzle_and_pack(struct vc4_compile *c, nir_builder *b,
nir_ssa_def **colors)
@@ -537,8 +489,6 @@ vc4_nir_blend_pipeline(struct vc4_compile *c, nir_builder *b, nir_ssa_def *src,
if (c->fs_key->sample_alpha_to_one && c->fs_key->msaa)
src_color[3] = nir_imm_float(b, 1.0);
- vc4_nir_emit_alpha_test_discard(c, b, src_color[3]);
-
nir_ssa_def *packed_color;
if (srgb) {
/* Unswizzle the destination color. */
diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index e93333d35e77..bf7424bf28a7 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -2250,8 +2250,15 @@ vc4_shader_ntq(struct vc4_context *vc4, enum qstage stage,
c->s = nir_shader_clone(c, key->shader_state->base.ir.nir);
- if (stage == QSTAGE_FRAG)
+ if (stage == QSTAGE_FRAG) {
+ if (c->fs_key->alpha_test_func != COMPARE_FUNC_ALWAYS) {
+ NIR_PASS_V(c->s, nir_lower_alpha_test,
+ c->fs_key->alpha_test_func,
+ c->fs_key->sample_alpha_to_one &&
+ c->fs_key->msaa);
+ }
NIR_PASS_V(c->s, vc4_nir_lower_blend, c);
+ }
struct nir_lower_tex_options tex_options = {
/* We would need to implement txs, but we don't want the
@@ -2748,10 +2755,10 @@ vc4_update_compiled_fs(struct vc4_context *vc4, uint8_t prim_mode)
key->stencil_full_writemasks = vc4->zsa->stencil_uniforms[2] != 0;
key->depth_enabled = (vc4->zsa->base.depth.enabled ||
key->stencil_enabled);
- if (vc4->zsa->base.alpha.enabled) {
- key->alpha_test = true;
+ if (vc4->zsa->base.alpha.enabled)
key->alpha_test_func = vc4->zsa->base.alpha.func;
- }
+ else
+ key->alpha_test_func = COMPARE_FUNC_ALWAYS;
if (key->is_points) {
key->point_sprite_mask =
diff --git a/src/gallium/drivers/vc4/vc4_qir.h b/src/gallium/drivers/vc4/vc4_qir.h
index 6469e51b051c..90acaef28989 100644
--- a/src/gallium/drivers/vc4/vc4_qir.h
+++ b/src/gallium/drivers/vc4/vc4_qir.h
@@ -354,7 +354,6 @@ struct vc4_fs_key {
bool stencil_full_writemasks;
bool is_points;
bool is_lines;
- bool alpha_test;
bool point_coord_upper_left;
bool light_twoside;
bool msaa;
--
2.14.1
More information about the mesa-dev
mailing list