radeonsi: support fragment shader per primitive input

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37932>

src/gallium/drivers/radeonsi/si_shader_info.h

@@ -206,6 +206,7 @@ struct si_shader_variant_info {
    uint32_t vs_output_ps_input_cntl[NUM_TOTAL_VARYING_SLOTS];
    union si_ps_input_info ps_inputs[SI_NUM_INTERP];
    uint8_t num_ps_inputs;
+   uint8_t num_ps_per_primitive_inputs;
    uint8_t ps_colors_read;
    uint8_t num_input_sgprs;
    uint8_t num_input_vgprs;

src/gallium/drivers/radeonsi/si_shader_variant_info.c

@@ -25,6 +25,9 @@ void si_get_shader_variant_info(struct si_shader *shader,
        */
       for (unsigned i = 0; i < ARRAY_SIZE(shader->info.ps_inputs); i++)
          shader->info.ps_inputs[i].interpolate = INTERP_MODE_FLAT;
+
+      shader->info.num_ps_per_primitive_inputs =
+         util_bitcount64(nir->info.per_primitive_inputs);
    }
 
    nir_foreach_block(block, nir_shader_get_entrypoint(nir)) {
@@ -50,6 +53,7 @@ void si_get_shader_variant_info(struct si_shader *shader,
             case nir_intrinsic_load_input:
             case nir_intrinsic_load_input_vertex:
             case nir_intrinsic_load_per_vertex_input:
+            case nir_intrinsic_load_per_primitive_input:
             case nir_intrinsic_load_interpolated_input: {
                if (nir->info.stage == MESA_SHADER_VERTEX) {
                   shader->info.uses_vmem_load_other = true;
@@ -78,6 +82,9 @@ void si_get_shader_variant_info(struct si_shader *shader,
                      shader->info.ps_inputs[index].interpolate = INTERP_MODE_SMOOTH;
                      if (intr->def.bit_size == 16)
                         shader->info.ps_inputs[index].fp16_lo_hi_valid |= 0x1 << sem.high_16bits;
+                  } else if (intr->intrinsic == nir_intrinsic_load_per_primitive_input) {
+                     /* per primitive input from mesh shader */
+                     shader->info.ps_inputs[index].interpolate = INTERP_MODE_NONE;
                   }
                }
                break;

src/gallium/drivers/radeonsi/si_state_shaders.cpp

@@ -2249,7 +2249,18 @@ static void si_shader_ps(struct si_screen *sscreen, struct si_shader *shader)
                        (sscreen->info.gfx_level == GFX11 && !shader->ps.num_interp &&
                         shader->config.lds_size);
 
-      shader->ps.spi_ps_in_control = S_0286D8_NUM_INTERP(shader->ps.num_interp) |
+      unsigned num_prim_interp = 0;
+      unsigned num_interp = shader->ps.num_interp;
+      if (sscreen->info.gfx_level == GFX10_3) {
+         /* NUM_INTERP / NUM_PRIM_INTERP separately contain
+          * the number of per-vertex and per-primitive PS input attributes.
+          */
+         num_prim_interp = shader->info.num_ps_per_primitive_inputs;
+         num_interp -= num_prim_interp;
+      }
+
+      shader->ps.spi_ps_in_control = S_0286D8_NUM_INTERP(num_interp) |
+                                     S_0286D8_NUM_PRIM_INTERP(num_prim_interp) |
                                      S_0286D8_PARAM_GEN(param_gen) |
                                      S_0286D8_PS_W32_EN(shader->wave_size == 32);
    }