vc4: Cook up the draw-time VPM setup info during shader compile.

This will give the compiler the chance to dead-code eliminate unused VPM
reads.  This is particularly a big deal in the CS where a bunch of vattrs
are just not going to be used.

4 files changed, 28 insertions, 11 deletions

diff --git a/src/gallium/drivers/vc4/vc4_context.h b/src/gallium/drivers/vc4/vc4_context.h
index 7e18a75e5b6..90a68e5c28e 100644
--- a/src/gallium/drivers/vc4/vc4_context.h
+++ b/src/gallium/drivers/vc4/vc4_context.h
@@ -121,6 +121,12 @@ struct vc4_compiled_shader {
 
         uint8_t num_inputs;
 
+        /* Byte offsets for the start of the vertex attributes 0-7, and the
+         * total size as "attribute" 8.
+         */
+        uint8_t vattr_offsets[9];
+        uint8_t vattrs_live;
+
         /**
          * Array of the meanings of the VPM inputs this shader needs.
          *
diff --git a/src/gallium/drivers/vc4/vc4_draw.c b/src/gallium/drivers/vc4/vc4_draw.c
index cc3f2d42183..bb4b9a42217 100644
--- a/src/gallium/drivers/vc4/vc4_draw.c
+++ b/src/gallium/drivers/vc4/vc4_draw.c
@@ -185,14 +185,14 @@ vc4_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
         cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
 
         cl_u16(&vc4->shader_rec, 0); /* vs num uniforms */
-        cl_u8(&vc4->shader_rec, (1 << num_elements_emit) - 1); /* vs attribute array bitfield */
-        cl_u8(&vc4->shader_rec, 16 * num_elements_emit); /* vs total attribute size */
+        cl_u8(&vc4->shader_rec, vc4->prog.vs->vattrs_live);
+        cl_u8(&vc4->shader_rec, vc4->prog.vs->vattr_offsets[8]);
         cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo, 0);
         cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
 
         cl_u16(&vc4->shader_rec, 0); /* cs num uniforms */
-        cl_u8(&vc4->shader_rec, (1 << num_elements_emit) - 1); /* cs attribute array bitfield */
-        cl_u8(&vc4->shader_rec, 16 * num_elements_emit); /* cs total attribute size */
+        cl_u8(&vc4->shader_rec, vc4->prog.cs->vattrs_live);
+        cl_u8(&vc4->shader_rec, vc4->prog.cs->vattr_offsets[8]);
         cl_reloc(vc4, &vc4->shader_rec, vc4->prog.cs->bo, 0);
         cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
 
@@ -211,8 +211,8 @@ vc4_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
                 cl_reloc(vc4, &vc4->shader_rec, rsc->bo, offset);
                 cl_u8(&vc4->shader_rec, elem_size - 1);
                 cl_u8(&vc4->shader_rec, vb->stride);
-                cl_u8(&vc4->shader_rec, vpm_offset); /* VS VPM offset */
-                cl_u8(&vc4->shader_rec, vpm_offset); /* CS VPM offset */
+                cl_u8(&vc4->shader_rec, vc4->prog.vs->vattr_offsets[i]);
+                cl_u8(&vc4->shader_rec, vc4->prog.cs->vattr_offsets[i]);
 
                 vpm_offset += align(elem_size, 4);
 
diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index 56cd5c27d7b..581b9400957 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -1079,6 +1079,7 @@ emit_vertex_input(struct vc4_compile *c, int attr)
         uint32_t attr_size = util_format_get_blocksize(format);
         struct qreg vpm_reads[4];
 
+        c->vattr_sizes[attr] = align(attr_size, 4);
         for (int i = 0; i < align(attr_size, 4) / 4; i++) {
                 struct qreg vpm = { QFILE_VPM, attr * 4 + i };
                 vpm_reads[i] = qir_MOV(c, vpm);
@@ -1933,11 +1934,10 @@ emit_stub_vpm_read(struct vc4_compile *c)
         if (c->num_inputs)
                 return;
 
-        for (int i = 0; i < 4; i++) {
-                struct qreg vpm = { QFILE_VPM, 0 };
-                (void)qir_MOV(c, vpm);
-                c->num_inputs++;
-        }
+        c->vattr_sizes[0] = 4;
+        struct qreg vpm = { QFILE_VPM, 0 };
+        (void)qir_MOV(c, vpm);
+        c->num_inputs++;
 }
 
 static void
@@ -2275,6 +2275,15 @@ vc4_get_compiled_shader(struct vc4_context *vc4, enum qstage stage,
                 }
         } else {
                 shader->num_inputs = c->num_inputs;
+
+                shader->vattr_offsets[0] = 0;
+                for (int i = 0; i < 8; i++) {
+                        shader->vattr_offsets[i + 1] =
+                                shader->vattr_offsets[i] + c->vattr_sizes[i];
+
+                        if (c->vattr_sizes[i])
+                                shader->vattrs_live |= (1 << i);
+                }
         }
 
         copy_uniform_state_to_shader(shader, c);
diff --git a/src/gallium/drivers/vc4/vc4_qir.h b/src/gallium/drivers/vc4/vc4_qir.h
index ebec7ccfbe3..d2f89ae9e69 100644
--- a/src/gallium/drivers/vc4/vc4_qir.h
+++ b/src/gallium/drivers/vc4/vc4_qir.h
@@ -312,6 +312,8 @@ struct vc4_compile {
         struct qreg line_x, point_x, point_y;
         struct qreg discard;
 
+        uint8_t vattr_sizes[8];
+
         /**
          * Array of the TGSI semantics of all FS QFILE_VARY reads.
          *