nv50: allow per-sample interpolation to be forced via rast

Uses the same technique as for nvc0 of fixups before upload, and
evicting in case of state change. Removes one source of variants kept by
st/mesa.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

8 files changed, 52 insertions, 9 deletions

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h b/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
index 391130e67af..c0cab3299b5 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
@@ -144,6 +144,7 @@ struct nv50_ir_prog_info
          bool earlyFragTests;
          bool separateFragData;
          bool usesDiscard;
+         bool sampleInterp;      /* perform sample interp on all fp inputs */
       } fp;
       struct {
          uint32_t inputOffset; /* base address for user args */
@@ -168,7 +169,6 @@ struct nv50_ir_prog_info
       int8_t viewportId;         /* output index of ViewportIndex */
       uint8_t fragDepth;         /* output index of FragDepth */
       uint8_t sampleMask;        /* output index of SampleMask */
-      bool sampleInterp;         /* perform sample interp on all fp inputs */
       uint8_t backFaceColor[2];  /* input/output indices of back face colour */
       uint8_t globalAccess;      /* 1 for read, 2 for wr, 3 for rw */
       bool fp64;                 /* program uses fp64 math */
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
index 90147668c91..e9b1cef2aa0 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
@@ -876,6 +876,30 @@ CodeEmitterNV50::emitPFETCH(const Instruction *i)
    emitFlagsRd(i);
 }
 
+static void
+interpApply(const InterpEntry *entry, uint32_t *code,
+      bool force_persample_interp, bool flatshade)
+{
+   int ipa = entry->ipa;
+   int encSize = entry->reg;
+   int loc = entry->loc;
+
+   if ((ipa & NV50_IR_INTERP_SAMPLE_MASK) == NV50_IR_INTERP_DEFAULT &&
+       (ipa & NV50_IR_INTERP_MODE_MASK) != NV50_IR_INTERP_FLAT) {
+      if (force_persample_interp) {
+         if (encSize == 8)
+            code[loc + 1] |= 1 << 16;
+         else
+            code[loc + 0] |= 1 << 24;
+      } else {
+         if (encSize == 8)
+            code[loc + 1] &= ~(1 << 16);
+         else
+            code[loc + 0] &= ~(1 << 24);
+      }
+   }
+}
+
 void
 CodeEmitterNV50::emitINTERP(const Instruction *i)
 {
@@ -904,6 +928,8 @@ CodeEmitterNV50::emitINTERP(const Instruction *i)
       code[0] |= 1;
       emitFlagsRd(i);
    }
+
+   addInterp(i->ipa, i->encSize, interpApply);
 }
 
 void
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
index eb28622eb59..0031b5e15e8 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -1125,7 +1125,7 @@ bool Source::scanDeclaration(const struct tgsi_full_declaration *decl)
          break;
       case TGSI_SEMANTIC_SAMPLEID:
       case TGSI_SEMANTIC_SAMPLEPOS:
-         info->io.sampleInterp = 1;
+         info->prop.fp.sampleInterp = 1;
          break;
       default:
          break;
@@ -1478,7 +1478,7 @@ Converter::translateInterpMode(const struct nv50_ir_varying *var, operation& op)
    op = (mode == NV50_IR_INTERP_PERSPECTIVE || mode == NV50_IR_INTERP_SC)
       ? OP_PINTERP : OP_LINTERP;
 
-   if (var->centroid || info->io.sampleInterp)
+   if (var->centroid || info->prop.fp.sampleInterp)
       mode |= NV50_IR_INTERP_CENTROID;
 
    return mode;
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_program.c b/src/gallium/drivers/nouveau/nv50/nv50_program.c
index eff4477472c..26ba2c2c380 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_program.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_program.c
@@ -336,7 +336,6 @@ nv50_program_translate(struct nv50_program *prog, uint16_t chipset)
    info->io.ucpCBSlot = 15;
    info->io.ucpBase = NV50_CB_AUX_UCP_OFFSET;
    info->io.genUserClip = prog->vp.clpd_nr;
-   info->io.sampleInterp = prog->fp.sample_interp;
 
    info->io.resInfoCBSlot = 15;
    info->io.suInfoBase = NV50_CB_AUX_TEX_MS_OFFSET;
@@ -374,6 +373,7 @@ nv50_program_translate(struct nv50_program *prog, uint16_t chipset)
    prog->code = info->bin.code;
    prog->code_size = info->bin.codeSize;
    prog->fixups = info->bin.relocData;
+   prog->interps = info->bin.interpData;
    prog->max_gpr = MAX2(4, (info->bin.maxGPR >> 1) + 1);
    prog->tls_space = info->bin.tlsSpace;
 
@@ -456,6 +456,10 @@ nv50_program_upload_code(struct nv50_context *nv50, struct nv50_program *prog)
 
    if (prog->fixups)
       nv50_ir_relocate_code(prog->fixups, prog->code, prog->code_base, 0, 0);
+   if (prog->interps)
+      nv50_ir_change_interp(prog->interps, prog->code,
+                            prog->fp.force_persample_interp,
+                            false /* flatshade */);
 
    nv50_sifc_linear_u8(&nv50->base, nv50->screen->code,
                        (prog->type << NV50_CODE_BO_SIZE_LOG2) + prog->code_base,
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_program.h b/src/gallium/drivers/nouveau/nv50/nv50_program.h
index f4e8e9402ca..24cc96567d7 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_program.h
+++ b/src/gallium/drivers/nouveau/nv50/nv50_program.h
@@ -86,7 +86,7 @@ struct nv50_program {
       uint32_t interp; /* 0x1988 */
       uint32_t colors; /* 0x1904 */
       uint8_t has_samplemask;
-      uint8_t sample_interp;
+      uint8_t force_persample_interp;
    } fp;
 
    struct {
@@ -99,6 +99,7 @@ struct nv50_program {
    } gp;
 
    void *fixups; /* relocation records */
+   void *interps; /* interpolation records */
 
    struct nouveau_heap *mem;
 
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_screen.c b/src/gallium/drivers/nouveau/nv50/nv50_screen.c
index bcad90128d2..398abfb97ba 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_screen.c
@@ -192,6 +192,7 @@ nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_INDEP_BLEND_FUNC:
    case PIPE_CAP_TEXTURE_QUERY_LOD:
    case PIPE_CAP_SAMPLE_SHADING:
+   case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
       return class_3d >= NVA3_3D_CLASS;
 
    /* unsupported caps */
@@ -216,7 +217,6 @@ nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
-   case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
    case PIPE_CAP_SHAREABLE_SHADERS:
       return 0;
 
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_shader_state.c b/src/gallium/drivers/nouveau/nv50/nv50_shader_state.c
index 941555ffbf8..b3fece07c3f 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_shader_state.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_shader_state.c
@@ -168,11 +168,23 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 {
    struct nouveau_pushbuf *push = nv50->base.pushbuf;
    struct nv50_program *fp = nv50->fragprog;
+   struct pipe_rasterizer_state *rast = &nv50->rast->pipe;
 
-   fp->fp.sample_interp = nv50->min_samples > 1;
+   if (fp->fp.force_persample_interp != rast->force_persample_interp) {
+      /* Force the program to be reuploaded, which will trigger interp fixups
+       * to get applied
+       */
+      if (fp->mem)
+         nouveau_heap_free(&fp->mem);
+
+      fp->fp.force_persample_interp = rast->force_persample_interp;
+   }
+
+   if (fp->mem && !(nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_MIN_SAMPLES)))
+      return;
 
    if (!nv50_program_validate(nv50, fp))
-         return;
+      return;
    nv50_program_update_context_state(nv50, fp, 1);
 
    BEGIN_NV04(push, NV50_3D(FP_REG_ALLOC_TEMP), 1);
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_state_validate.c b/src/gallium/drivers/nouveau/nv50/nv50_state_validate.c
index 66dcf43533b..b6181edf24f 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_state_validate.c
@@ -487,7 +487,7 @@ static struct state_validate {
     { nv50_validate_viewport,      NV50_NEW_VIEWPORT },
     { nv50_vertprog_validate,      NV50_NEW_VERTPROG },
     { nv50_gmtyprog_validate,      NV50_NEW_GMTYPROG },
-    { nv50_fragprog_validate,      NV50_NEW_FRAGPROG |
+    { nv50_fragprog_validate,      NV50_NEW_FRAGPROG | NV50_NEW_RASTERIZER |
                                    NV50_NEW_MIN_SAMPLES },
     { nv50_fp_linkage_validate,    NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG |
                                    NV50_NEW_GMTYPROG | NV50_NEW_RASTERIZER },