freedreno/a5xx+: Set the IJ_LINEAR_* request bits if we need the regs.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12949>

diff --git a/src/freedreno/vulkan/tu_pipeline.c b/src/freedreno/vulkan/tu_pipeline.c
index 0813484..4b9b7bb 100644 (file)
--- a/src/freedreno/vulkan/tu_pipeline.c
+++ b/src/freedreno/vulkan/tu_pipeline.c
@@ -1364,12 +1364,6 @@ tu6_emit_fs_inputs(struct tu_cs *cs, const struct ir3_shader_variant *fs)
    for (unsigned i = 0; i < ARRAY_SIZE(ij_regid); i++)
       ij_regid[i] = ir3_find_sysval_regid(fs, SYSTEM_VALUE_BARYCENTRIC_PERSP_PIXEL + i);
 
-   if (VALIDREG(ij_regid[IJ_LINEAR_SAMPLE]))
-      tu_finishme("linear sample varying");
-
-   if (VALIDREG(ij_regid[IJ_LINEAR_CENTROID]))
-      tu_finishme("linear centroid varying");
-
    if (fs->num_sampler_prefetch > 0) {
       assert(VALIDREG(ij_regid[IJ_PERSP_PIXEL]));
       /* also, it seems like ij_pix is *required* to be r0.x */
@@ -1430,14 +1424,15 @@ tu6_emit_fs_inputs(struct tu_cs *cs, const struct ir3_shader_variant *fs)
       else
          need_size = true;
    }
-   if (VALIDREG(ij_regid[IJ_LINEAR_PIXEL]))
-      need_size = true;
 
    tu_cs_emit_pkt4(cs, REG_A6XX_GRAS_CNTL, 1);
    tu_cs_emit(cs,
          CONDREG(ij_regid[IJ_PERSP_PIXEL], A6XX_GRAS_CNTL_IJ_PERSP_PIXEL) |
          CONDREG(ij_regid[IJ_PERSP_CENTROID], A6XX_GRAS_CNTL_IJ_PERSP_CENTROID) |
          CONDREG(ij_regid[IJ_PERSP_SAMPLE], A6XX_GRAS_CNTL_IJ_PERSP_SAMPLE) |
+         CONDREG(ij_regid[IJ_LINEAR_PIXEL], A6XX_GRAS_CNTL_IJ_LINEAR_PIXEL) |
+         CONDREG(ij_regid[IJ_LINEAR_CENTROID], A6XX_GRAS_CNTL_IJ_LINEAR_CENTROID) |
+         CONDREG(ij_regid[IJ_LINEAR_SAMPLE], A6XX_GRAS_CNTL_IJ_LINEAR_SAMPLE) |
          COND(need_size, A6XX_GRAS_CNTL_IJ_LINEAR_PIXEL) |
          COND(need_size_persamp, A6XX_GRAS_CNTL_IJ_LINEAR_SAMPLE) |
          COND(fs->fragcoord_compmask != 0, A6XX_GRAS_CNTL_COORD_MASK(fs->fragcoord_compmask)));
@@ -1447,6 +1442,9 @@ tu6_emit_fs_inputs(struct tu_cs *cs, const struct ir3_shader_variant *fs)
          CONDREG(ij_regid[IJ_PERSP_PIXEL], A6XX_RB_RENDER_CONTROL0_IJ_PERSP_PIXEL) |
          CONDREG(ij_regid[IJ_PERSP_CENTROID], A6XX_RB_RENDER_CONTROL0_IJ_PERSP_CENTROID) |
          CONDREG(ij_regid[IJ_PERSP_SAMPLE], A6XX_RB_RENDER_CONTROL0_IJ_PERSP_SAMPLE) |
+         CONDREG(ij_regid[IJ_LINEAR_PIXEL], A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_PIXEL) |
+         CONDREG(ij_regid[IJ_LINEAR_CENTROID], A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_CENTROID) |
+         CONDREG(ij_regid[IJ_LINEAR_SAMPLE], A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_SAMPLE) |
          COND(need_size, A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_PIXEL) |
          COND(enable_varyings, A6XX_RB_RENDER_CONTROL0_UNK10) |
          COND(need_size_persamp, A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_SAMPLE) |

diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_program.c b/src/gallium/drivers/freedreno/a5xx/fd5_program.c
index 531c4e2..2236420 100644 (file)
--- a/src/gallium/drivers/freedreno/a5xx/fd5_program.c
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_program.c
@@ -542,15 +542,18 @@ fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
    OUT_PKT4(ring, REG_A5XX_SP_SP_CNTL, 1);
    OUT_RING(ring, 0x00000010); /* XXX */
 
-   /* XXX: missing enable bits for per-sample bary linear centroid and
-    * IJ_PERSP_SIZE (should be identical to a6xx)
-    */
-
    OUT_PKT4(ring, REG_A5XX_GRAS_CNTL, 1);
    OUT_RING(ring,
             CONDREG(ij_regid[IJ_PERSP_PIXEL], A5XX_GRAS_CNTL_IJ_PERSP_PIXEL) |
                CONDREG(ij_regid[IJ_PERSP_CENTROID],
                        A5XX_GRAS_CNTL_IJ_PERSP_CENTROID) |
+               CONDREG(ij_regid[IJ_PERSP_SAMPLE],
+                       A5XX_GRAS_CNTL_IJ_PERSP_SAMPLE) |
+               CONDREG(ij_regid[IJ_LINEAR_PIXEL], A5XX_GRAS_CNTL_IJ_LINEAR_PIXEL) |
+               CONDREG(ij_regid[IJ_LINEAR_CENTROID],
+                       A5XX_GRAS_CNTL_IJ_LINEAR_CENTROID) |
+               CONDREG(ij_regid[IJ_LINEAR_SAMPLE],
+                       A5XX_GRAS_CNTL_IJ_LINEAR_SAMPLE) |
                COND(s[FS].v->fragcoord_compmask != 0,
                     A5XX_GRAS_CNTL_COORD_MASK(s[FS].v->fragcoord_compmask) |
                        A5XX_GRAS_CNTL_IJ_LINEAR_PIXEL) |
@@ -564,6 +567,14 @@ fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
               A5XX_RB_RENDER_CONTROL0_IJ_PERSP_PIXEL) |
          CONDREG(ij_regid[IJ_PERSP_CENTROID],
                  A5XX_RB_RENDER_CONTROL0_IJ_PERSP_CENTROID) |
+         CONDREG(ij_regid[IJ_PERSP_SAMPLE],
+                 A5XX_RB_RENDER_CONTROL0_IJ_PERSP_SAMPLE) |
+         CONDREG(ij_regid[IJ_LINEAR_PIXEL],
+              A5XX_RB_RENDER_CONTROL0_IJ_LINEAR_PIXEL) |
+         CONDREG(ij_regid[IJ_LINEAR_CENTROID],
+                 A5XX_RB_RENDER_CONTROL0_IJ_LINEAR_CENTROID) |
+         CONDREG(ij_regid[IJ_LINEAR_SAMPLE],
+                 A5XX_RB_RENDER_CONTROL0_IJ_LINEAR_SAMPLE) |
          COND(s[FS].v->fragcoord_compmask != 0,
               A5XX_RB_RENDER_CONTROL0_COORD_MASK(s[FS].v->fragcoord_compmask) |
                  A5XX_RB_RENDER_CONTROL0_IJ_LINEAR_PIXEL) |

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_program.c b/src/gallium/drivers/freedreno/a6xx/fd6_program.c
index c4d9137..621cce2 100644 (file)
--- a/src/gallium/drivers/freedreno/a6xx/fd6_program.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_program.c
@@ -834,10 +834,6 @@ setup_stateobj(struct fd_ringbuffer *ring, struct fd_context *ctx,
       else
          need_size = true;
    }
-   if (VALIDREG(ij_regid[IJ_LINEAR_PIXEL]))
-      need_size = true;
-
-   /* XXX: enable bits for linear centroid and linear sample bary */
 
    OUT_PKT4(ring, REG_A6XX_GRAS_CNTL, 1);
    OUT_RING(
@@ -846,6 +842,10 @@ setup_stateobj(struct fd_ringbuffer *ring, struct fd_context *ctx,
          CONDREG(ij_regid[IJ_PERSP_CENTROID],
                  A6XX_GRAS_CNTL_IJ_PERSP_CENTROID) |
          CONDREG(ij_regid[IJ_PERSP_SAMPLE], A6XX_GRAS_CNTL_IJ_PERSP_SAMPLE) |
+         CONDREG(ij_regid[IJ_LINEAR_PIXEL], A6XX_GRAS_CNTL_IJ_LINEAR_PIXEL) |
+         CONDREG(ij_regid[IJ_LINEAR_CENTROID],
+                 A6XX_GRAS_CNTL_IJ_LINEAR_CENTROID) |
+         CONDREG(ij_regid[IJ_LINEAR_SAMPLE], A6XX_GRAS_CNTL_IJ_LINEAR_SAMPLE) |
          COND(need_size, A6XX_GRAS_CNTL_IJ_LINEAR_PIXEL) |
          COND(need_size_persamp, A6XX_GRAS_CNTL_IJ_LINEAR_SAMPLE) |
          COND(fs->fragcoord_compmask != 0,
@@ -860,6 +860,12 @@ setup_stateobj(struct fd_ringbuffer *ring, struct fd_context *ctx,
                  A6XX_RB_RENDER_CONTROL0_IJ_PERSP_CENTROID) |
          CONDREG(ij_regid[IJ_PERSP_SAMPLE],
                  A6XX_RB_RENDER_CONTROL0_IJ_PERSP_SAMPLE) |
+         CONDREG(ij_regid[IJ_LINEAR_PIXEL],
+              A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_PIXEL) |
+         CONDREG(ij_regid[IJ_LINEAR_CENTROID],
+                 A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_CENTROID) |
+         CONDREG(ij_regid[IJ_LINEAR_SAMPLE],
+                 A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_SAMPLE) |
          COND(need_size, A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_PIXEL) |
          COND(enable_varyings, A6XX_RB_RENDER_CONTROL0_UNK10) |
          COND(need_size_persamp, A6XX_RB_RENDER_CONTROL0_IJ_LINEAR_SAMPLE) |