broadcom/vc5: Move texture return channel setup into the compiler.

The compiler decides how many LDTMUs we're going to emit, and that must
match the P1 flags.  This brings the return channel counting to a single
place (so all that's passed into the compiler is "how many return channels
you may request from this texture's format), and was a necessary step for
shadow samplers once we stop using OVRTMUOUT=0.

src/broadcom/cle/v3d_packet_v33.xml

@@ -810,10 +810,7 @@
 
   <struct name="Texture Uniform Parameter 1 CFG_MODE=1">
     <field name="Texture state record base address" size="28" start="4" type="address"/>
-    <field name="Return word 3 of texture data" size="1" start="3" type="bool"/>
-    <field name="Return word 2 of texture data" size="1" start="2" type="bool"/>
-    <field name="Return word 1 of texture data" size="1" start="1" type="bool"/>
-    <field name="Return word 0 of texture data" size="1" start="0" type="bool"/>
+    <field name="Return words of texture data" size="4" start="0" type="uint"/>
   </struct>
 
   <struct name="Texture Shader State">

src/broadcom/compiler/nir_to_vir.c

@@ -337,6 +337,9 @@ ntq_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
                 .fetch_sample_mode = instr->op == nir_texop_txf,
         };
 
+        struct V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1 p1_unpacked = {
+        };
+
         switch (instr->sampler_dim) {
         case GLSL_SAMPLER_DIM_1D:
                 if (instr->is_array)
@@ -419,11 +422,35 @@ ntq_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
                 }
         }
 
+        bool return_16 = (c->key->tex[unit].return_size == 16 ||
+                          p0_unpacked.shadow);
+
+        /* Limit the number of channels returned to both how many the NIR
+         * instruction writes and how many the instruction could produce.
+         */
+        uint32_t instr_return_channels = nir_tex_instr_dest_size(instr);
+        if (return_16)
+                instr_return_channels = (instr_return_channels + 1) / 2;
+
+        p1_unpacked.return_words_of_texture_data =
+                (1 << MIN2(instr_return_channels,
+                           c->key->tex[unit].return_channels)) - 1;
+
         uint32_t p0_packed;
         V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1_pack(NULL,
                                                          (uint8_t *)&p0_packed,
                                                          &p0_unpacked);
 
+        uint32_t p1_packed;
+        V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1_pack(NULL,
+                                                         (uint8_t *)&p1_packed,
+                                                         &p1_unpacked);
+        /* Load unit number into the address field, which will be be used by
+         * the driver to decide which texture to put in the actual address
+         * field.
+         */
+        p1_packed |= unit << 5;
+
         /* There is no native support for GL texture rectangle coordinates, so
          * we have to rescale from ([0, width], [0, height]) to ([0, 1], [0,
          * 1]).
@@ -439,7 +466,7 @@ ntq_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
 
         struct qreg texture_u[] = {
                 vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P0_0 + unit, p0_packed),
-                vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P1, unit),
+                vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P1, p1_packed),
         };
         uint32_t next_texture_u = 0;
 
@@ -460,17 +487,16 @@ ntq_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
                 }
         }
 
-        bool return_16 = (c->key->tex[unit].return_size == 16 ||
-                          p0_unpacked.shadow);
-
         struct qreg return_values[4];
-        for (int i = 0; i < c->key->tex[unit].return_channels; i++)
-                return_values[i] = vir_LDTMU(c);
-        /* Swizzling .zw of an RG texture should give undefined results, not
-         * crash the compiler.
-         */
-        for (int i = c->key->tex[unit].return_channels; i < 4; i++)
-                return_values[i] = c->undef;
+        for (int i = 0; i < 4; i++) {
+                /* Swizzling .zw of an RG texture should give undefined
+                 * results, not crash the compiler.
+                 */
+                if (p1_unpacked.return_words_of_texture_data & (1 << i))
+                        return_values[i] = vir_LDTMU(c);
+                else
+                        return_values[i] = c->undef;
+        }
 
         for (int i = 0; i < nir_tex_instr_dest_size(instr); i++) {
                 struct qreg chan;

src/broadcom/compiler/v3d_compiler.h

@@ -215,9 +215,9 @@ enum quniform_contents {
          * A reference to a texture config parameter 1 uniform.
          *
          * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
-         * defines texture width, height, filters, and wrap modes.  It will be
-         * found as a parameter to the second QOP_TEX_[STRB] instruction in a
-         * sequence.
+         * has the pointer to the indirect texture state.  Our data[] field
+         * will have a packed p1 value, but the address field will be just
+         * which texture unit's texture should be referenced.
          */
         QUNIFORM_TEXTURE_CONFIG_P1,