v3d: Add a lowering pass for line smoothing

When line smoothing is enabled, the driver now increases the width of
the line so that it can add some semi-transparent pixels to either side
of the line. A lowering pass is added which modifies the alpha component
of every write to fragment output 0 so that if the fragment is outside
the width of the line then the alpha is reduced. It additionally
discards fragments that are completely invisible. It might seem bad to
use discard on a tiled renderer but the assumption is that any bad
effects from using discard will also happen anyway because of enabling
alpha blending.

v2: Disable the line smoothing pass entirely when the framebuffer
    contains an integer colour output or one with no alpha channel.
    Calculate the coverage once upfront and store in a global variable
    instead of calculating each time an output write is modified. Also
    do the conditional discard once upfront.
v3: Don’t check whether the output buffer has an alpha channel. Only
    look at output 0. Use aa_line_width intrinsic instead of calculating
    the real line width in the shader. Clamp the coverage as part of the
    global variable, not per output write.

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5624>

src/broadcom/Makefile.sources

@@ -41,6 +41,7 @@ BROADCOM_FILES = \
 	compiler/v3d_compiler.h \
 	compiler/v3d_nir_lower_image_load_store.c \
 	compiler/v3d_nir_lower_io.c \
+	compiler/v3d_nir_lower_line_smooth.c \
 	compiler/v3d_nir_lower_scratch.c \
 	compiler/v3d_nir_lower_txf_ms.c \
 	qpu/qpu_disasm.c \

src/broadcom/compiler/meson.build

@@ -37,6 +37,7 @@ libbroadcom_compiler_files = files(
   'v3d_compiler.h',
   'v3d_nir_lower_io.c',
   'v3d_nir_lower_image_load_store.c',
+  'v3d_nir_lower_line_smooth.c',
   'v3d_nir_lower_logic_ops.c',
   'v3d_nir_lower_scratch.c',
   'v3d_nir_lower_txf_ms.c',

src/broadcom/compiler/v3d_compiler.h

@@ -352,6 +352,7 @@ struct v3d_fs_key {
         bool depth_enabled;
         bool is_points;
         bool is_lines;
+        bool line_smoothing;
         bool alpha_test;
         bool point_coord_upper_left;
         bool light_twoside;
@@ -867,6 +868,7 @@ bool vir_opt_small_immediates(struct v3d_compile *c);
 bool vir_opt_vpm(struct v3d_compile *c);
 void v3d_nir_lower_blend(nir_shader *s, struct v3d_compile *c);
 void v3d_nir_lower_io(nir_shader *s, struct v3d_compile *c);
+void v3d_nir_lower_line_smooth(nir_shader *shader);
 void v3d_nir_lower_logic_ops(nir_shader *s, struct v3d_compile *c);
 void v3d_nir_lower_scratch(nir_shader *s);
 void v3d_nir_lower_txf_ms(nir_shader *s, struct v3d_compile *c);

src/broadcom/compiler/v3d_nir_lower_line_smooth.c

@@ -0,0 +1,163 @@
+/*
+ * Copyright © 2020 Raspberry Pi
+ *
+ * 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.
+ */
+
+#include "compiler/v3d_compiler.h"
+#include "compiler/nir/nir_builder.h"
+#include <math.h>
+
+/**
+ * Lowers line smoothing by modifying the alpha component of fragment outputs
+ * using the distance from the center of the line.
+ */
+
+struct lower_line_smooth_state {
+        nir_shader *shader;
+        nir_variable *coverage;
+};
+
+static void
+lower_line_smooth_intrinsic(struct lower_line_smooth_state *state,
+                            nir_builder *b,
+                            nir_intrinsic_instr *intr)
+{
+        b->cursor = nir_before_instr(&intr->instr);
+
+        nir_ssa_def *one = nir_imm_float(b, 1.0f);
+
+        nir_ssa_def *coverage = nir_load_var(b, state->coverage);
+
+        nir_ssa_def *new_val = nir_fmul(b, nir_vec4(b, one, one, one, coverage),
+                                        intr->src[0].ssa);
+
+        nir_instr_rewrite_src(&intr->instr,
+                              &intr->src[0],
+                              nir_src_for_ssa(new_val));
+}
+
+static void
+lower_line_smooth_func(struct lower_line_smooth_state *state,
+                       nir_function_impl *impl)
+{
+        nir_builder b;
+
+        nir_builder_init(&b, impl);
+
+        nir_foreach_block(block, impl) {
+                nir_foreach_instr_safe(instr, block) {
+                        if (instr->type != nir_instr_type_intrinsic)
+                                continue;
+
+                        nir_intrinsic_instr *intr =
+                                nir_instr_as_intrinsic(instr);
+
+                        if (intr->intrinsic != nir_intrinsic_store_output ||
+                            nir_intrinsic_base(intr) != 0 ||
+                            intr->num_components != 4 ||
+                            !intr->src[0].is_ssa)
+                                continue;
+
+                        lower_line_smooth_intrinsic(state, &b, intr);
+                }
+        }
+}
+
+static void
+initialise_coverage_var(struct lower_line_smooth_state *state,
+                        nir_function_impl *impl)
+{
+        nir_builder b;
+
+        nir_builder_init(&b, impl);
+
+        b.cursor = nir_before_block(nir_start_block(impl));
+
+        nir_ssa_def *line_width = nir_load_line_width(&b);
+
+        nir_ssa_def *real_line_width = nir_load_aa_line_width(&b);
+
+        /* The line coord varies from 0.0 to 1.0 across the width of the line */
+        nir_ssa_def *line_coord = nir_load_line_coord(&b);
+
+        /* fabs(line_coord - 0.5) * real_line_width */
+        nir_ssa_def *pixels_from_center =
+                nir_fmul(&b, real_line_width,
+                         nir_fabs(&b, nir_fsub(&b, line_coord,
+                                               nir_imm_float(&b, 0.5f))));
+
+        /* 0.5 - 1/√2 * (pixels_from_center - line_width * 0.5) */
+        nir_ssa_def *coverage =
+                nir_fsub(&b,
+                         nir_imm_float(&b, 0.5f),
+                         nir_fmul(&b,
+                                  nir_imm_float(&b, 1.0f / M_SQRT2),
+                                  nir_fsub(&b, pixels_from_center,
+                                           nir_fmul(&b,
+                                                    line_width,
+                                                    nir_imm_float(&b, 0.5f)))));
+
+        /* Discard fragments that aren’t covered at all by the line */
+        nir_ssa_def *outside = nir_fge(&b, nir_imm_float(&b, 0.0f), coverage);
+
+        nir_intrinsic_instr *discard =
+                nir_intrinsic_instr_create(state->shader,
+                                           nir_intrinsic_discard_if);
+        discard->src[0] = nir_src_for_ssa(outside);
+        nir_builder_instr_insert(&b, &discard->instr);
+
+        /* Clamp to at most 1.0. If it was less than 0.0 then the fragment will
+         * be discarded so we don’t need to handle that.
+         */
+        nir_ssa_def *clamped = nir_fmin(&b, coverage, nir_imm_float(&b, 1.0f));
+
+        nir_store_var(&b, state->coverage, clamped, 0x1 /* writemask */);
+}
+
+static nir_variable *
+make_coverage_var(nir_shader *s)
+{
+        nir_variable *var = nir_variable_create(s,
+                                                nir_var_shader_temp,
+                                                glsl_float_type(),
+                                                "line_coverage");
+        var->data.how_declared = nir_var_hidden;
+
+        return var;
+}
+
+void
+v3d_nir_lower_line_smooth(nir_shader *s)
+{
+        assert(s->info.stage == MESA_SHADER_FRAGMENT);
+
+        struct lower_line_smooth_state state = {
+                .shader = s,
+                .coverage = make_coverage_var(s),
+        };
+
+        nir_foreach_function(function, s) {
+                if (function->is_entrypoint)
+                        initialise_coverage_var(&state, function->impl);
+
+                lower_line_smooth_func(&state, function->impl);
+        }
+}

src/broadcom/compiler/vir.c

@@ -890,6 +890,13 @@ v3d_nir_lower_fs_early(struct v3d_compile *c)
 
         NIR_PASS_V(c->s, v3d_nir_lower_logic_ops, c);
 
+        if (c->fs_key->line_smoothing) {
+                v3d_nir_lower_line_smooth(c->s);
+                NIR_PASS_V(c->s, nir_lower_global_vars_to_local);
+                /* The lowering pass can introduce new sysval reads */
+                nir_shader_gather_info(c->s, nir_shader_get_entrypoint(c->s));
+        }
+
         /* If the shader has no non-TLB side effects, we can promote it to
          * enabling early_fragment_tests even if the user didn't.
          */

src/gallium/drivers/v3d/v3d_context.c

@@ -149,6 +149,51 @@ v3d_update_primitive_counters(struct v3d_context *v3d)
         }
 }
 
+bool
+v3d_line_smoothing_enabled(struct v3d_context *v3d)
+{
+        if (!v3d->rasterizer->base.line_smooth)
+                return false;
+
+        /* According to the OpenGL docs, line smoothing shouldn’t be applied
+         * when multisampling
+         */
+        if (v3d->job->msaa || v3d->rasterizer->base.multisample)
+                return false;
+
+        if (v3d->framebuffer.nr_cbufs <= 0)
+                return false;
+
+        struct pipe_surface *cbuf = v3d->framebuffer.cbufs[0];
+        if (!cbuf)
+                return false;
+
+        /* Modifying the alpha for pure integer formats probably
+         * doesn’t make sense because we don’t know how the application
+         * uses the alpha value.
+         */
+        if (util_format_is_pure_integer(cbuf->format))
+                return false;
+
+        return true;
+}
+
+float
+v3d_get_real_line_width(struct v3d_context *v3d)
+{
+        float width = v3d->rasterizer->base.line_width;
+
+        if (v3d_line_smoothing_enabled(v3d)) {
+                /* If line smoothing is enabled then we want to add some extra
+                 * pixels to the width in order to have some semi-transparent
+                 * edges.
+                 */
+                width = floorf(M_SQRT2 * width) + 3;
+        }
+
+        return width;
+}
+
 static void
 v3d_context_destroy(struct pipe_context *pctx)
 {

src/gallium/drivers/v3d/v3d_context.h

@@ -699,6 +699,10 @@ struct v3d_fence *v3d_fence_create(struct v3d_context *v3d);
 
 void v3d_update_primitive_counters(struct v3d_context *v3d);
 
+bool v3d_line_smoothing_enabled(struct v3d_context *v3d);
+
+float v3d_get_real_line_width(struct v3d_context *v3d);
+
 #ifdef v3dX
 #  include "v3dx_context.h"
 #else

src/gallium/drivers/v3d/v3d_program.c

@@ -541,6 +541,8 @@ v3d_update_compiled_fs(struct v3d_context *v3d, uint8_t prim_mode)
         key->is_points = (prim_mode == PIPE_PRIM_POINTS);
         key->is_lines = (prim_mode >= PIPE_PRIM_LINES &&
                          prim_mode <= PIPE_PRIM_LINE_STRIP);
+        key->line_smoothing = (key->is_lines &&
+                               v3d_line_smoothing_enabled(v3d));
         key->clamp_color = v3d->rasterizer->base.clamp_fragment_color;
         if (v3d->blend->base.logicop_enable) {
                 key->logicop_func = v3d->blend->base.logicop_func;

src/gallium/drivers/v3d/v3d_uniforms.c

@@ -313,11 +313,14 @@ v3d_write_uniforms(struct v3d_context *v3d, struct v3d_job *job,
                         break;
 
                 case QUNIFORM_LINE_WIDTH:
-                case QUNIFORM_AA_LINE_WIDTH:
                         cl_aligned_f(&uniforms,
                                      v3d->rasterizer->base.line_width);
                         break;
 
+                case QUNIFORM_AA_LINE_WIDTH:
+                        cl_aligned_f(&uniforms, v3d_get_real_line_width(v3d));
+                        break;
+
                 case QUNIFORM_UBO_ADDR: {
                         uint32_t unit = v3d_unit_data_get_unit(data);
                         /* Constant buffer 0 may be a system memory pointer,

src/gallium/drivers/v3d/v3dx_emit.c

@@ -551,7 +551,7 @@ v3dX(emit_state)(struct pipe_context *pctx)
                 }
 
                 cl_emit(&job->bcl, LINE_WIDTH, line_width) {
-                        line_width.line_width = v3d->rasterizer->base.line_width;
+                        line_width.line_width = v3d_get_real_line_width(v3d);
                 }
         }