broadcom/vc5: Start hooking up multiple render targets support.

We now emit as many TLB color writes as there are color buffers.

diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index acfeba4cb715edd4041cdf155ed0492c720eb099..642173e3ad37c279e340643b80c468cca9f30d78 100644 (file)
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -1070,24 +1070,26 @@ emit_frag_end(struct v3d_compile *c)
          * uniform setup
          */
 
-        if (c->output_color_var) {
-                nir_variable *var = c->output_color_var;
+        for (int rt = 0; rt < c->fs_key->nr_cbufs; rt++) {
+                if (!c->output_color_var[rt])
+                        continue;
+
+                nir_variable *var = c->output_color_var[rt];
                 struct qreg *color = &c->outputs[var->data.driver_location * 4];
                 int num_components = glsl_get_vector_elements(var->type);
-                uint32_t conf = ~0;
+                uint32_t conf = 0xffffff00;
                 struct qinst *inst;
 
+                conf |= TLB_SAMPLE_MODE_PER_PIXEL;
+                conf |= (7 - rt) << TLB_RENDER_TARGET_SHIFT;
+
                 assert(num_components != 0);
                 switch (glsl_get_base_type(var->type)) {
                 case GLSL_TYPE_UINT:
                 case GLSL_TYPE_INT:
-                        conf = (TLB_TYPE_I32_COLOR |
-                                TLB_SAMPLE_MODE_PER_PIXEL |
-                                ((7 - 0) << TLB_RENDER_TARGET_SHIFT) |
-                                ((num_components - 1) <<
-                                 TLB_VEC_SIZE_MINUS_1_SHIFT) |
-                                0xffffff00);
-
+                        conf |= TLB_TYPE_I32_COLOR;
+                        conf |= ((num_components - 1) <<
+                                 TLB_VEC_SIZE_MINUS_1_SHIFT);
 
                         inst = vir_MOV_dest(c, vir_reg(QFILE_TLBU, 0), color[0]);
                         vir_set_cond(inst, discard_cond);
@@ -1107,13 +1109,26 @@ emit_frag_end(struct v3d_compile *c)
                         struct qreg b = color[2];
                         struct qreg a = color[3];
 
-                        if (c->fs_key->swap_color_rb)  {
+                        conf |= TLB_TYPE_F16_COLOR;
+                        conf |= TLB_F16_SWAP_HI_LO;
+                        if (num_components >= 3)
+                                conf |= TLB_VEC_SIZE_4_F16;
+                        else
+                                conf |= TLB_VEC_SIZE_2_F16;
+
+                        if (c->fs_key->swap_color_rb & (1 << rt))  {
                                 r = color[2];
                                 b = color[0];
                         }
 
                         inst = vir_VFPACK_dest(c, vir_reg(QFILE_TLB, 0), r, g);
+                        if (conf != ~0) {
+                                inst->dst.file = QFILE_TLBU;
+                                inst->src[vir_get_implicit_uniform_src(inst)] =
+                                        vir_uniform_ui(c, conf);
+                        }
                         vir_set_cond(inst, discard_cond);
+
                         inst = vir_VFPACK_dest(c, vir_reg(QFILE_TLB, 0), b, a);
                         vir_set_cond(inst, discard_cond);
                         break;
@@ -1421,8 +1436,17 @@ ntq_setup_outputs(struct v3d_compile *c)
                 if (c->s->stage == MESA_SHADER_FRAGMENT) {
                         switch (var->data.location) {
                         case FRAG_RESULT_COLOR:
+                                c->output_color_var[0] = var;
+                                c->output_color_var[1] = var;
+                                c->output_color_var[2] = var;
+                                c->output_color_var[3] = var;
+                                break;
                         case FRAG_RESULT_DATA0:
-                                c->output_color_var = var;
+                        case FRAG_RESULT_DATA1:
+                        case FRAG_RESULT_DATA2:
+                        case FRAG_RESULT_DATA3:
+                                c->output_color_var[var->data.location -
+                                                    FRAG_RESULT_DATA0] = var;
                                 break;
                         case FRAG_RESULT_DEPTH:
                                 c->output_position_index = loc;

diff --git a/src/broadcom/compiler/v3d_compiler.h b/src/broadcom/compiler/v3d_compiler.h
index a29b08ac1ab47e4614a91419a2703df5bd1df132..bdd505ee9ba7e85b29fabbdc02c1e74927f2b0af 100644 (file)
--- a/src/broadcom/compiler/v3d_compiler.h
+++ b/src/broadcom/compiler/v3d_compiler.h
@@ -318,7 +318,8 @@ struct v3d_fs_key {
         bool sample_alpha_to_coverage;
         bool sample_alpha_to_one;
         bool clamp_color;
-        bool swap_color_rb;
+        uint8_t nr_cbufs;
+        uint8_t swap_color_rb;
         uint8_t alpha_test_func;
         uint8_t logicop_func;
         uint32_t point_sprite_mask;
@@ -490,7 +491,7 @@ struct v3d_compile {
         uint32_t num_uniforms;
         uint32_t num_outputs;
         uint32_t output_position_index;
-        nir_variable *output_color_var;
+        nir_variable *output_color_var[4];
         uint32_t output_point_size_index;
         uint32_t output_sample_mask_index;
 

diff --git a/src/gallium/drivers/vc5/vc5_program.c b/src/gallium/drivers/vc5/vc5_program.c
index f22f474954cdd919354842eacb958ecf62bcdd8b..d5634c7a7d3fdfb885f8819ed56359941ada4a1e 100644 (file)
--- a/src/gallium/drivers/vc5/vc5_program.c
+++ b/src/gallium/drivers/vc5/vc5_program.c
@@ -357,12 +357,18 @@ vc5_update_compiled_fs(struct vc5_context *vc5, uint8_t prim_mode)
                 key->alpha_test_func = vc5->zsa->base.alpha.func;
         }
 
-        if (vc5->framebuffer.cbufs[0]) {
-                struct pipe_surface *cbuf = vc5->framebuffer.cbufs[0];
+        /* gl_FragColor's propagation to however many bound color buffers
+         * there are means that the buffer count needs to be in the key.
+         */
+        key->nr_cbufs = vc5->framebuffer.nr_cbufs;
+
+        for (int i = 0; i < key->nr_cbufs; i++) {
+                struct pipe_surface *cbuf = vc5->framebuffer.cbufs[i];
                 const struct util_format_description *desc =
                         util_format_description(cbuf->format);
 
-                key->swap_color_rb = desc->swizzle[0] == PIPE_SWIZZLE_Z;
+                if (desc->swizzle[0] == PIPE_SWIZZLE_Z)
+                        key->swap_color_rb |= 1 << i;
         }
 
         if (key->is_points) {