radeonsi: Enable nir_lower_point_smooth lowering pass for point smoothing

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15117>

diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 80e0367..9984045 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -1137,6 +1137,9 @@ struct si_context {
    bool ps_uses_fbfetch;
    bool smoothing_enabled;
 
+   /* point smoothing state.*/
+   bool point_smoothing_enabled;
+
    /* DB render state. */
    unsigned ps_db_shader_control;
    unsigned dbcb_copy_sample;

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 9874c96..531fc60 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -43,6 +43,18 @@ static const char scratch_rsrc_dword1_symbol[] = "SCRATCH_RSRC_DWORD1";
 
 static void si_dump_shader_key(const struct si_shader *shader, FILE *f);
 
+/* Get the number of all interpolated inputs */
+unsigned si_get_ps_num_interp(struct si_shader *ps)
+{
+   struct si_shader_info *info = &ps->selector->info;
+   unsigned num_colors = !!(info->colors_read & 0x0f) + !!(info->colors_read & 0xf0);
+   unsigned num_interp =
+      ps->selector->info.num_inputs + (ps->key.ps.part.prolog.color_two_side ? num_colors : 0);
+
+   assert(num_interp <= 32);
+   return MIN2(num_interp, 32);
+}
+
 /** Whether the shader runs as a combination of multiple API shaders */
 bool si_is_multi_part_shader(struct si_shader *shader)
 {
@@ -1293,6 +1305,7 @@ static void si_dump_shader_key(const struct si_shader *shader, FILE *f)
       fprintf(f, "  epilog.clamp_color = %u\n", key->ps.part.epilog.clamp_color);
       fprintf(f, "  epilog.dual_src_blend_swizzle = %u\n", key->ps.part.epilog.dual_src_blend_swizzle);
       fprintf(f, "  mono.poly_line_smoothing = %u\n", key->ps.mono.poly_line_smoothing);
+      fprintf(f, "  mono.point_smoothing = %u\n", key->ps.mono.point_smoothing);
       fprintf(f, "  mono.interpolate_at_sample_force_center = %u\n",
               key->ps.mono.interpolate_at_sample_force_center);
       fprintf(f, "  mono.fbfetch_msaa = %u\n", key->ps.mono.fbfetch_msaa);
@@ -1650,6 +1663,9 @@ struct nir_shader *si_get_nir_shader(struct si_shader *shader, bool *free_nir,
    if (sel->stage == MESA_SHADER_FRAGMENT && key->ps.mono.poly_line_smoothing)
       NIR_PASS(progress, nir, nir_lower_poly_line_smooth, SI_NUM_SMOOTH_AA_SAMPLES);
 
+   if (sel->stage == MESA_SHADER_FRAGMENT && key->ps.mono.point_smoothing)
+      NIR_PASS(progress, nir, nir_lower_point_smooth);
+
    if (progress)
       si_nir_opts(sel->screen, nir, true);
 

diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 8bed98d..dff39ae 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -733,6 +733,7 @@ struct si_shader_key_ps {
    /* Flags for monolithic compilation only. */
    struct {
       unsigned poly_line_smoothing : 1;
+      unsigned point_smoothing : 1;
       unsigned interpolate_at_sample_force_center : 1;
       unsigned fbfetch_msaa : 1;
       unsigned fbfetch_is_1D : 1;
@@ -979,6 +980,7 @@ void si_multiwave_lds_size_workaround(struct si_screen *sscreen, unsigned *lds_s
 const char *si_get_shader_name(const struct si_shader *shader);
 void si_shader_binary_clean(struct si_shader_binary *binary);
 struct nir_shader *si_deserialize_shader(struct si_shader_selector *sel);
+unsigned si_get_ps_num_interp(struct si_shader *ps);
 
 /* si_shader_info.c */
 void si_nir_scan_shader(struct si_screen *sscreen,  const struct nir_shader *nir,
@@ -1059,6 +1061,7 @@ static inline bool si_shader_uses_discard(struct si_shader *shader)
    /* Changes to this should also update ps_modifies_zs. */
    return shader->selector->info.base.fs.uses_discard ||
           shader->key.ps.part.prolog.poly_stipple ||
+          shader->key.ps.mono.point_smoothing ||
           shader->key.ps.part.epilog.alpha_func != PIPE_FUNC_ALWAYS;
 }
 

diff --git a/src/gallium/drivers/radeonsi/si_shader_llvm.c b/src/gallium/drivers/radeonsi/si_shader_llvm.c
index b6023f4..fc3f932 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_llvm.c
+++ b/src/gallium/drivers/radeonsi/si_shader_llvm.c
@@ -899,6 +899,7 @@ bool si_llvm_translate_nir(struct si_shader_context *ctx, struct si_shader *shad
          ctx->abi.color1 = ac_to_integer(&ctx->ac, ac_build_gather_values(&ctx->ac, values, 4));
       }
 
+      ctx->abi.num_interp = si_get_ps_num_interp(shader);
       ctx->abi.interp_at_sample_force_center =
          ctx->shader->key.ps.mono.interpolate_at_sample_force_center;
 

diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 94c9241..be87e24 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -949,6 +949,7 @@ static void *si_create_rs_state(struct pipe_context *ctx, const struct pipe_rast
    rs->line_smooth = state->line_smooth;
    rs->line_width = state->line_width;
    rs->poly_smooth = state->poly_smooth;
+   rs->point_smooth = state->point_smooth;
    rs->uses_poly_offset = state->offset_point || state->offset_line || state->offset_tri;
    rs->clamp_fragment_color = state->clamp_fragment_color;
    rs->clamp_vertex_color = state->clamp_vertex_color;
@@ -1195,6 +1196,7 @@ static void si_bind_rs_state(struct pipe_context *ctx, void *state)
        old_rs->multisample_enable != rs->multisample_enable ||
        old_rs->poly_stipple_enable != rs->poly_stipple_enable ||
        old_rs->poly_smooth != rs->poly_smooth || old_rs->line_smooth != rs->line_smooth ||
+       old_rs->point_smooth != rs->point_smooth ||
        old_rs->clamp_fragment_color != rs->clamp_fragment_color ||
        old_rs->force_persample_interp != rs->force_persample_interp ||
        old_rs->polygon_mode_is_points != rs->polygon_mode_is_points) {
@@ -1207,6 +1209,7 @@ static void si_bind_rs_state(struct pipe_context *ctx, void *state)
 
    if (old_rs->line_smooth != rs->line_smooth ||
        old_rs->poly_smooth != rs->poly_smooth ||
+       old_rs->point_smooth != rs->point_smooth ||
        old_rs->poly_stipple_enable != rs->poly_stipple_enable ||
        old_rs->flatshade != rs->flatshade)
       si_update_vrs_flat_shading(sctx);

diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index e48e166..4ca3595 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -91,6 +91,7 @@ struct si_state_rasterizer {
    unsigned poly_stipple_enable : 1;
    unsigned line_smooth : 1;
    unsigned poly_smooth : 1;
+   unsigned point_smooth : 1;
    unsigned uses_poly_offset : 1;
    unsigned clamp_fragment_color : 1;
    unsigned clamp_vertex_color : 1;

diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.cpp b/src/gallium/drivers/radeonsi/si_state_shaders.cpp
index fa349a2..722e5d4 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state_shaders.cpp
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.cpp
@@ -1770,17 +1770,6 @@ static void si_shader_vs(struct si_screen *sscreen, struct si_shader *shader,
    polaris_set_vgt_vertex_reuse(sscreen, shader->selector, shader);
 }
 
-static unsigned si_get_ps_num_interp(struct si_shader *ps)
-{
-   struct si_shader_info *info = &ps->selector->info;
-   unsigned num_colors = !!(info->colors_read & 0x0f) + !!(info->colors_read & 0xf0);
-   unsigned num_interp =
-      ps->selector->info.num_inputs + (ps->key.ps.part.prolog.color_two_side ? num_colors : 0);
-
-   assert(num_interp <= 32);
-   return MIN2(num_interp, 32);
-}
-
 static unsigned si_get_spi_shader_col_format(struct si_shader *shader)
 {
    unsigned spi_shader_col_format = shader->key.ps.part.epilog.spi_shader_col_format;
@@ -1994,8 +1983,11 @@ static void si_shader_ps(struct si_screen *sscreen, struct si_shader *shader)
    spi_ps_in_control = S_0286D8_NUM_INTERP(num_interp) |
                        S_0286D8_PS_W32_EN(shader->wave_size == 32);
 
-   /* Workaround when there are no PS inputs but LDS is used. */
-   if (sscreen->info.gfx_level == GFX11 && !num_interp && shader->config.lds_size)
+   /* Enable PARAM_GEN for point smoothing.
+    * Gfx11 workaround when there are no PS inputs but LDS is used.
+    */
+   if ((sscreen->info.gfx_level == GFX11 && !num_interp && shader->config.lds_size) ||
+       shader->key.ps.mono.point_smoothing)
       spi_ps_in_control |= S_0286D8_PARAM_GEN(1);
 
    shader->ctx_reg.ps.num_interp = num_interp;
@@ -2187,7 +2179,8 @@ void si_update_ps_inputs_read_or_disabled(struct si_context *sctx)
                             ps->info.writes_samplemask ||
                             sctx->queued.named.blend->alpha_to_coverage ||
                             sctx->queued.named.dsa->alpha_func != PIPE_FUNC_ALWAYS ||
-                            sctx->queued.named.rasterizer->poly_stipple_enable;
+                            sctx->queued.named.rasterizer->poly_stipple_enable ||
+                            sctx->queued.named.rasterizer->point_smooth;
       unsigned ps_colormask = si_get_total_colormask(sctx);
 
       ps_disabled = sctx->queued.named.rasterizer->rasterizer_discard ||
@@ -2388,6 +2381,9 @@ static void si_ps_key_update_primtype_shader_rasterizer_framebuffer(struct si_co
    key->ps.mono.poly_line_smoothing =
       ((is_poly && rs->poly_smooth) || (is_line && rs->line_smooth)) &&
       sctx->framebuffer.nr_samples <= 1;
+
+   key->ps.mono.point_smoothing = rs->point_smooth &&
+                                  sctx->current_rast_prim == PIPE_PRIM_POINTS;
 }
 
 void si_ps_key_update_sample_shading(struct si_context *sctx)
@@ -3552,7 +3548,7 @@ void si_update_vrs_flat_shading(struct si_context *sctx)
 
       if (allow_flat_shading &&
           (rs->line_smooth || rs->poly_smooth || rs->poly_stipple_enable ||
-           (!rs->flatshade && info->uses_interp_color)))
+           rs->point_smooth || (!rs->flatshade && info->uses_interp_color)))
          allow_flat_shading = false;
 
       if (sctx->allow_flat_shading != allow_flat_shading) {