nvc0: fix up image support for allowing multiple samples

Basically we just have to scale up the coordinates and then add the
relevant sample offset. The code to handle this was already largely
present from Christoph's earlier attempts to pipe images through back in
the dark ages, this just hooks it all up.

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

7 files changed, 108 insertions, 49 deletions

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 ed3249e67b1..41dc18f8859 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -2388,6 +2388,9 @@ Converter::getImageCoords(std::vector<Value *> &coords, int r, int s)
 
    for (int c = 0; c < arg; ++c)
       coords.push_back(fetchSrc(s, c));
+
+   if (t.isMS())
+      coords.push_back(fetchSrc(s, 3));
 }
 
 // For raw loads, granularity is 4 byte.
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
index 67bd73bc8c7..0558ae0e7c0 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -1569,44 +1569,51 @@ static inline uint16_t getSuClampSubOp(const TexInstruction *su, int c)
 bool
 NVC0LoweringPass::handleSUQ(TexInstruction *suq)
 {
+   int mask = suq->tex.mask;
    int dim = suq->tex.target.getDim();
    int arg = dim + (suq->tex.target.isArray() || suq->tex.target.isCube());
-   uint8_t s = prog->driver->io.auxCBSlot;
    Value *ind = suq->getIndirectR();
    uint32_t base;
-   int c;
-
-   base = prog->driver->io.suInfoBase + suq->tex.r * NVE4_SU_INFO__STRIDE;
+   int c, d;
 
-   if (ind)
-      ind = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getScratch(),
+   if (ind) {
+      ind = bld.mkOp2v(OP_ADD, TYPE_U32, bld.getSSA(),
+                       ind, bld.mkImm(suq->tex.r));
+      ind = bld.mkOp2v(OP_AND, TYPE_U32, bld.getSSA(),
+                       ind, bld.mkImm(7));
+      ind = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getSSA(),
                        ind, bld.mkImm(6));
+      base = 0;
+   } else {
+      base = suq->tex.r * NVE4_SU_INFO__STRIDE;
+   }
 
-   for (c = 0; c < arg; ++c) {
-      if (suq->defExists(c)) {
-         int offset;
+   for (c = 0, d = 0; c < 3; ++c, mask >>= 1) {
+      if (c >= arg || !(mask & 1))
+         continue;
 
-         if (c == 1 && suq->tex.target == TEX_TARGET_1D_ARRAY) {
-            offset = base + NVE4_SU_INFO_SIZE(2);
-         } else {
-            offset = base + NVE4_SU_INFO_SIZE(c);
-         }
-         bld.mkLoad(TYPE_U32, suq->getDef(c),
-                    bld.mkSymbol(FILE_MEMORY_CONST, s, TYPE_U32, offset), ind);
-      }
-   }
+      int offset;
 
-   if (suq->tex.target.isCube()) {
-      if (suq->defExists(2)) {
-         bld.mkOp2(OP_DIV, TYPE_U32, suq->getDef(2), suq->getDef(2),
-                   bld.loadImm(NULL, 6));
+      if (c == 1 && suq->tex.target == TEX_TARGET_1D_ARRAY) {
+         offset = NVE4_SU_INFO_SIZE(2);
+      } else {
+         offset = NVE4_SU_INFO_SIZE(c);
       }
+      bld.mkMov(suq->getDef(d++), loadSuInfo32(ind, base + offset));
+      if (c == 2 && suq->tex.target.isCube())
+         bld.mkOp2(OP_DIV, TYPE_U32, suq->getDef(d - 1), suq->getDef(d - 1),
+                   bld.loadImm(NULL, 6));
    }
 
-   if (suq->defExists(3)) {
-      // .w contains the number of samples for multi-sampled images but we
-      // don't support them for now.
-      bld.mkMov(suq->getDef(3), bld.loadImm(NULL, 1));
+   if (mask & 1) {
+      if (suq->tex.target.isMS()) {
+         Value *ms_x = loadSuInfo32(ind, base + NVE4_SU_INFO_MS(0));
+         Value *ms_y = loadSuInfo32(ind, base + NVE4_SU_INFO_MS(1));
+         Value *ms = bld.mkOp2v(OP_ADD, TYPE_U32, bld.getScratch(), ms_x, ms_y);
+         bld.mkOp2(OP_SHL, TYPE_U32, suq->getDef(d++), bld.loadImm(NULL, 1), ms);
+      } else {
+         bld.mkMov(suq->getDef(d++), bld.loadImm(NULL, 1));
+      }
    }
 
    bld.remove(suq);
@@ -1616,7 +1623,7 @@ NVC0LoweringPass::handleSUQ(TexInstruction *suq)
 void
 NVC0LoweringPass::adjustCoordinatesMS(TexInstruction *tex)
 {
-   const uint16_t base = tex->tex.r * NVE4_SU_INFO__STRIDE;
+   uint16_t base;
    const int arg = tex->tex.target.getArgCount();
 
    if (tex->tex.target == TEX_TARGET_2D_MS)
@@ -1632,13 +1639,18 @@ NVC0LoweringPass::adjustCoordinatesMS(TexInstruction *tex)
    Value *s = tex->getSrc(arg - 1);
 
    Value *tx = bld.getSSA(), *ty = bld.getSSA(), *ts = bld.getSSA();
-   Value *ind = NULL;
+   Value *ind = tex->getIndirectR();
 
-   if (tex->tex.rIndirectSrc >= 0) {
-      assert(tex->tex.r == 0);
-      // FIXME: out of bounds
+   if (ind) {
+      ind = bld.mkOp2v(OP_ADD, TYPE_U32, bld.getSSA(),
+                       ind, bld.mkImm(tex->tex.r));
+      ind = bld.mkOp2v(OP_AND, TYPE_U32, bld.getSSA(),
+                       ind, bld.mkImm(7));
       ind = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getSSA(),
-                       tex->getIndirectR(), bld.mkImm(6));
+                       ind, bld.mkImm(6));
+      base = 0;
+   } else {
+      base = tex->tex.r * NVE4_SU_INFO__STRIDE;
    }
 
    Value *ms_x = loadSuInfo32(ind, base + NVE4_SU_INFO_MS(0));
@@ -2044,6 +2056,10 @@ NVC0LoweringPass::processSurfaceCoordsNVC0(TexInstruction *su)
    Value *v;
    Value *ind = NULL;
 
+   bld.setPosition(su, false);
+
+   adjustCoordinatesMS(su);
+
    if (su->tex.rIndirectSrc >= 0) {
       ind = su->getIndirectR();
       if (su->tex.r > 0) {
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_compute.c b/src/gallium/drivers/nouveau/nvc0/nvc0_compute.c
index 59bbe1efe10..66cae1916e1 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_compute.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_compute.c
@@ -113,6 +113,30 @@ nvc0_screen_compute_setup(struct nvc0_screen *screen,
    PUSH_DATA (push, screen->txc->offset + 65536);
    PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);
 
+   /* MS sample coordinate offsets */
+   BEGIN_NVC0(push, NVC0_CP(CB_SIZE), 3);
+   PUSH_DATA (push, 2048);
+   PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(5));
+   PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(5));
+   BEGIN_1IC0(push, NVC0_CP(CB_POS), 1 + 2 * 8);
+   PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
+   PUSH_DATA (push, 0); /* 0 */
+   PUSH_DATA (push, 0);
+   PUSH_DATA (push, 1); /* 1 */
+   PUSH_DATA (push, 0);
+   PUSH_DATA (push, 0); /* 2 */
+   PUSH_DATA (push, 1);
+   PUSH_DATA (push, 1); /* 3 */
+   PUSH_DATA (push, 1);
+   PUSH_DATA (push, 2); /* 4 */
+   PUSH_DATA (push, 0);
+   PUSH_DATA (push, 3); /* 5 */
+   PUSH_DATA (push, 0);
+   PUSH_DATA (push, 2); /* 6 */
+   PUSH_DATA (push, 1);
+   PUSH_DATA (push, 3); /* 7 */
+   PUSH_DATA (push, 1);
+
    return 0;
 }
 
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_context.h b/src/gallium/drivers/nouveau/nvc0/nvc0_context.h
index fe9f9f56e0a..098d2a1add0 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_context.h
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_context.h
@@ -112,7 +112,7 @@
 #define NVC0_CB_AUX_TEX_INFO(i)     0x020 + (i) * 4
 #define NVC0_CB_AUX_TEX_SIZE        (32 * 4)
 /* 8 sets of 32-bits coordinate offsets */
-#define NVC0_CB_AUX_MS_INFO         0x0a0 /* CP */
+#define NVC0_CB_AUX_MS_INFO         0x0a0
 #define NVC0_CB_AUX_MS_SIZE         (8 * 2 * 4)
 /* block/grid size, at 3 32-bits integers each and gridid */
 #define NVC0_CB_AUX_GRID_INFO       0x0e0 /* CP */
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_program.c b/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
index aba95114eb7..d75b702c809 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
@@ -555,29 +555,25 @@ nvc0_program_translate(struct nvc0_program *prog, uint16_t chipset,
 
    info->io.genUserClip = prog->vp.num_ucps;
    info->io.auxCBSlot = 15;
+   info->io.msInfoCBSlot = 15;
    info->io.ucpBase = NVC0_CB_AUX_UCP_INFO;
    info->io.drawInfoBase = NVC0_CB_AUX_DRAW_INFO;
+   info->io.msInfoBase = NVC0_CB_AUX_MS_INFO;
+   info->io.bufInfoBase = NVC0_CB_AUX_BUF_INFO(0);
+   info->io.suInfoBase = NVC0_CB_AUX_SU_INFO(0);
+   if (chipset >= NVISA_GK104_CHIPSET) {
+      info->io.texBindBase = NVC0_CB_AUX_TEX_INFO(0);
+   }
 
    if (prog->type == PIPE_SHADER_COMPUTE) {
       if (chipset >= NVISA_GK104_CHIPSET) {
          info->io.auxCBSlot = 7;
-         info->io.texBindBase = NVC0_CB_AUX_TEX_INFO(0);
+         info->io.msInfoCBSlot = 7;
          info->prop.cp.gridInfoBase = NVC0_CB_AUX_GRID_INFO;
          info->io.uboInfoBase = NVC0_CB_AUX_UBO_INFO(0);
       }
-      info->io.msInfoCBSlot = 0;
-      info->io.msInfoBase = NVC0_CB_AUX_MS_INFO;
-      info->io.bufInfoBase = NVC0_CB_AUX_BUF_INFO(0);
-      info->io.suInfoBase = NVC0_CB_AUX_SU_INFO(0);
    } else {
-      if (chipset >= NVISA_GK104_CHIPSET) {
-         info->io.texBindBase = NVC0_CB_AUX_TEX_INFO(0);
-      }
       info->io.sampleInfoBase = NVC0_CB_AUX_SAMPLE_INFO;
-      info->io.bufInfoBase = NVC0_CB_AUX_BUF_INFO(0);
-      info->io.suInfoBase = NVC0_CB_AUX_SU_INFO(0);
-      info->io.msInfoCBSlot = 15;
-      info->io.msInfoBase = 0; /* TODO */
    }
 
    info->assignSlots = nvc0_program_assign_varying_slots;
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
index abcbfbfbac4..e0bfd3b4144 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
@@ -975,6 +975,26 @@ nvc0_screen_create(struct nouveau_device *dev)
          BEGIN_NVC0(push, NVC0_3D(TEX_LIMITS(i)), 1);
          PUSH_DATA (push, 0x54);
       }
+
+      /* MS sample coordinate offsets: these do not work with _ALT modes ! */
+      BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 2 * 8);
+      PUSH_DATA (push, NVC0_CB_AUX_MS_INFO);
+      PUSH_DATA (push, 0); /* 0 */
+      PUSH_DATA (push, 0);
+      PUSH_DATA (push, 1); /* 1 */
+      PUSH_DATA (push, 0);
+      PUSH_DATA (push, 0); /* 2 */
+      PUSH_DATA (push, 1);
+      PUSH_DATA (push, 1); /* 3 */
+      PUSH_DATA (push, 1);
+      PUSH_DATA (push, 2); /* 4 */
+      PUSH_DATA (push, 0);
+      PUSH_DATA (push, 3); /* 5 */
+      PUSH_DATA (push, 0);
+      PUSH_DATA (push, 2); /* 6 */
+      PUSH_DATA (push, 1);
+      PUSH_DATA (push, 3); /* 7 */
+      PUSH_DATA (push, 1);
    }
    BEGIN_NVC0(push, NVC0_3D(LINKED_TSC), 1);
    PUSH_DATA (push, 0);
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_tex.c b/src/gallium/drivers/nouveau/nvc0/nvc0_tex.c
index 1a5d8ecbf16..21d8e9d6797 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_tex.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_tex.c
@@ -928,11 +928,11 @@ nve4_set_surface_info(struct nouveau_pushbuf *push,
       address += lvl->offset;
 
       info[0]  = address >> 8;
-      info[2]  = width - 1;
+      info[2]  = (width << mt->ms_x) - 1;
       /* NOTE: this is really important: */
       info[2] |= (0xff & nve4_su_format_aux_map[view->format]) << 22;
       info[3]  = (0x88 << 24) | (lvl->pitch / 64);
-      info[4]  = height - 1;
+      info[4]  = (height << mt->ms_y) - 1;
       info[4] |= (lvl->tile_mode & 0x0f0) << 25;
       info[4] |= NVC0_TILE_SHIFT_Y(lvl->tile_mode) << 22;
       info[5]  = mt->layer_stride >> 8;
@@ -1051,8 +1051,8 @@ nvc0_validate_suf(struct nvc0_context *nvc0, int s)
 
             PUSH_DATAh(push, address);
             PUSH_DATA (push, address);
-            PUSH_DATA (push, width);
-            PUSH_DATA (push, height);
+            PUSH_DATA (push, width << mt->ms_x);
+            PUSH_DATA (push, height << mt->ms_y);
             PUSH_DATA (push, rt);
             PUSH_DATA (push, lvl->tile_mode & 0xff); /* mask out z-tiling */
          }