broadcom/vc5: Add support for f32 render targets.

The TLB write code is getting ugly and needs a refactoring (that will
hopefully handle TLBU uniform coalescing as well).

diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index 642173e3ad37c279e340643b80c468cca9f30d78..675cad2c32f47b4b11418c74666b45f6b3e3a6b7 100644 (file)
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -1109,28 +1109,47 @@ emit_frag_end(struct v3d_compile *c)
                         struct qreg b = color[2];
                         struct qreg a = color[3];
 
-                        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->f32_color_rb) {
+                                conf |= TLB_TYPE_F32_COLOR;
+                                conf |= ((num_components - 1) <<
+                                         TLB_VEC_SIZE_MINUS_1_SHIFT);
+                        } else {
+                                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;
+                        if (c->fs_key->f32_color_rb & (1 << rt)) {
+                                inst = vir_MOV_dest(c, vir_reg(QFILE_TLBU, 0), color[0]);
+                                vir_set_cond(inst, discard_cond);
                                 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);
+                                for (int i = 1; i < num_components; i++) {
+                                        inst = vir_MOV_dest(c, vir_reg(QFILE_TLB, 0),
+                                                            color[i]);
+                                        vir_set_cond(inst, discard_cond);
+                                }
+                        } else {
+                                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;
                 }
                 }

diff --git a/src/broadcom/compiler/v3d_compiler.h b/src/broadcom/compiler/v3d_compiler.h
index bdd505ee9ba7e85b29fabbdc02c1e74927f2b0af..08db4fb0e2bb7df8edc86c10e83d356d704c7c63 100644 (file)
--- a/src/broadcom/compiler/v3d_compiler.h
+++ b/src/broadcom/compiler/v3d_compiler.h
@@ -320,6 +320,8 @@ struct v3d_fs_key {
         bool clamp_color;
         uint8_t nr_cbufs;
         uint8_t swap_color_rb;
+        /* Mask of which render targets need to be written as 32-bit floats */
+        uint8_t f32_color_rb;
         uint8_t alpha_test_func;
         uint8_t logicop_func;
         uint32_t point_sprite_mask;

diff --git a/src/gallium/drivers/vc5/vc5_program.c b/src/gallium/drivers/vc5/vc5_program.c
index d5634c7a7d3fdfb885f8819ed56359941ada4a1e..80865a25a4d2a41093215d51b282f54c93995f9d 100644 (file)
--- a/src/gallium/drivers/vc5/vc5_program.c
+++ b/src/gallium/drivers/vc5/vc5_program.c
@@ -369,6 +369,10 @@ vc5_update_compiled_fs(struct vc5_context *vc5, uint8_t prim_mode)
 
                 if (desc->swizzle[0] == PIPE_SWIZZLE_Z)
                         key->swap_color_rb |= 1 << i;
+                if (desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT &&
+                    desc->channel[0].size == 32) {
+                        key->f32_color_rb |= 1 << i;
+                }
         }
 
         if (key->is_points) {