vc4: Add a helper for making driver-specific NIR load_uniform for GL state

In order to move more of our lowering into NIR, we need the ability to
reference various pipeline state (like texture rectangle scaling factors
or blend colors), so we just set those up as a load_uniform with a big
offset to indicate that it's not within the shader's uniform storage and
is one of our state values.

2 files changed, 30 insertions, 2 deletions

diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index 4a3a277a9fb..fb1726c0d1e 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -33,8 +33,9 @@
 #include "tgsi/tgsi_info.h"
 #include "tgsi/tgsi_lowering.h"
 #include "tgsi/tgsi_parse.h"
+#include "glsl/nir/nir.h"
+#include "glsl/nir/nir_builder.h"
 #include "nir/tgsi_to_nir.h"
-
 #include "vc4_context.h"
 #include "vc4_qpu.h"
 #include "vc4_qir.h"
@@ -109,6 +110,19 @@ indirect_uniform_load(struct vc4_compile *c, nir_intrinsic_instr *intr)
         return qir_TEX_RESULT(c);
 }
 
+nir_ssa_def *vc4_nir_get_state_uniform(struct nir_builder *b,
+                                       enum quniform_contents contents)
+{
+        nir_intrinsic_instr *intr =
+                nir_intrinsic_instr_create(b->shader,
+                                           nir_intrinsic_load_uniform);
+        intr->const_index[0] = VC4_NIR_STATE_UNIFORM_OFFSET + contents;
+        intr->num_components = 1;
+        nir_ssa_dest_init(&intr->instr, &intr->dest, 1, NULL);
+        nir_builder_instr_insert(b, &intr->instr);
+        return &intr->dest.ssa;
+}
+
 static struct qreg *
 ntq_init_ssa_def(struct vc4_compile *c, nir_ssa_def *def)
 {
@@ -1808,7 +1822,14 @@ ntq_emit_intrinsic(struct vc4_compile *c, nir_intrinsic_instr *instr)
         switch (instr->intrinsic) {
         case nir_intrinsic_load_uniform:
                 assert(instr->num_components == 1);
-                *dest = qir_uniform(c, QUNIFORM_UNIFORM, instr->const_index[0]);
+                if (instr->const_index[0] < VC4_NIR_STATE_UNIFORM_OFFSET) {
+                        *dest = qir_uniform(c, QUNIFORM_UNIFORM,
+                                            instr->const_index[0]);
+                } else {
+                        *dest = qir_uniform(c, instr->const_index[0] -
+                                            VC4_NIR_STATE_UNIFORM_OFFSET,
+                                            0);
+                }
                 break;
 
         case nir_intrinsic_load_uniform_indirect:
diff --git a/src/gallium/drivers/vc4/vc4_qir.h b/src/gallium/drivers/vc4/vc4_qir.h
index 7a74018d9af..57e25de1b94 100644
--- a/src/gallium/drivers/vc4/vc4_qir.h
+++ b/src/gallium/drivers/vc4/vc4_qir.h
@@ -414,6 +414,11 @@ struct vc4_compile {
         uint32_t variant_id;
 };
 
+/* Special offset for nir_load_uniform values to get a QUNIFORM_*
+ * state-dependent value.
+ */
+#define VC4_NIR_STATE_UNIFORM_OFFSET		2000000000
+
 struct vc4_compile *qir_compile_init(void);
 void qir_compile_destroy(struct vc4_compile *c);
 struct qinst *qir_inst(enum qop op, struct qreg dst,
@@ -454,6 +459,8 @@ bool qir_opt_dead_code(struct vc4_compile *c);
 bool qir_opt_small_immediates(struct vc4_compile *c);
 bool qir_opt_vpm_writes(struct vc4_compile *c);
 void vc4_nir_lower_io(struct vc4_compile *c);
+nir_ssa_def *vc4_nir_get_state_uniform(struct nir_builder *b,
+                                       enum quniform_contents contents);
 void qir_lower_uniforms(struct vc4_compile *c);
 
 void qpu_schedule_instructions(struct vc4_compile *c);