nv50: increase size of shader code bo

512 KiB should be quite enough, but dynamic resize might be nicer.

4 files changed, 23 insertions, 14 deletions

diff --git a/src/gallium/drivers/nv50/nv50_pc_emit.c b/src/gallium/drivers/nv50/nv50_pc_emit.c
index f37dc51e6aa..252c58dd8ff 100644
--- a/src/gallium/drivers/nv50/nv50_pc_emit.c
+++ b/src/gallium/drivers/nv50/nv50_pc_emit.c
@@ -762,7 +762,8 @@ emit_flow(struct nv_pc *pc, struct nv_instruction *i, ubyte flow_op)
       new_fixup(pc, NV50_FIXUP_CODE_RELOC, 0, pos, 0xffff << 11, 9);
       new_fixup(pc, NV50_FIXUP_CODE_RELOC, 1, pos, 0x3f << 14, -4);
 
-      pc->emit[0] |= (pos / 4) << 11;
+      pc->emit[0] |= ((pos >>  2) & 0xffff) << 11;
+      pc->emit[1] |= ((pos >> 18) & 0x003f) << 14;
    }
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index 3f148436e81..3ccaff2838a 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -286,7 +286,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
    uint32_t tesla_class;
    unsigned stack_size, max_warps, tls_space;
    int ret;
-   unsigned i;
+   unsigned i, base;
 
    screen = CALLOC_STRUCT(nv50_screen);
    if (!screen)
@@ -425,25 +425,28 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
    BEGIN_RING(chan, RING_3D(ZCULL_REGION), 1); /* deactivate ZCULL */
    OUT_RING  (chan, 0x3f);
 
-   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16, 3 << 16, &screen->code);
+   ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 1 << 16,
+                        3 << NV50_CODE_BO_SIZE_LOG2, &screen->code);
    if (ret)
       goto fail;
 
-   nouveau_resource_init(&screen->vp_code_heap, 0, 1 << 16);
-   nouveau_resource_init(&screen->gp_code_heap, 0, 1 << 16);
-   nouveau_resource_init(&screen->fp_code_heap, 0, 1 << 16);
+   nouveau_resource_init(&screen->vp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
+   nouveau_resource_init(&screen->gp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
+   nouveau_resource_init(&screen->fp_code_heap, 0, 1 << NV50_CODE_BO_SIZE_LOG2);
+
+   base = 1 << NV50_CODE_BO_SIZE_LOG2;
 
    BEGIN_RING(chan, RING_3D(VP_ADDRESS_HIGH), 2);
-   OUT_RELOCh(chan, screen->code, 0 << 16, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
-   OUT_RELOCl(chan, screen->code, 0 << 16, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+   OUT_RELOCh(chan, screen->code, base * 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+   OUT_RELOCl(chan, screen->code, base * 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
 
    BEGIN_RING(chan, RING_3D(FP_ADDRESS_HIGH), 2);
-   OUT_RELOCh(chan, screen->code, 1 << 16, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
-   OUT_RELOCl(chan, screen->code, 1 << 16, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+   OUT_RELOCh(chan, screen->code, base * 1, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+   OUT_RELOCl(chan, screen->code, base * 1, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
 
    BEGIN_RING(chan, RING_3D(GP_ADDRESS_HIGH), 2);
-   OUT_RELOCh(chan, screen->code, 2 << 16, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
-   OUT_RELOCl(chan, screen->code, 2 << 16, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+   OUT_RELOCh(chan, screen->code, base * 2, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+   OUT_RELOCl(chan, screen->code, base * 2, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
 
    nouveau_device_get_param(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
 
diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h
index 3886d8068cc..aea434b8679 100644
--- a/src/gallium/drivers/nv50/nv50_screen.h
+++ b/src/gallium/drivers/nv50/nv50_screen.h
@@ -14,6 +14,8 @@
 
 struct nv50_context;
 
+#define NV50_CODE_BO_SIZE_LOG2 19
+
 #define NV50_SCRATCH_SIZE (2 << 20)
 #define NV50_SCRATCH_NR_BUFFERS 2
 
diff --git a/src/gallium/drivers/nv50/nv50_shader_state.c b/src/gallium/drivers/nv50/nv50_shader_state.c
index f7d78a3b11b..bea9c095bb3 100644
--- a/src/gallium/drivers/nv50/nv50_shader_state.c
+++ b/src/gallium/drivers/nv50/nv50_shader_state.c
@@ -138,6 +138,7 @@ nv50_program_validate(struct nv50_context *nv50, struct nv50_program *prog)
       return FALSE;
 
    if (prog->type == PIPE_SHADER_FRAGMENT) heap = nv50->screen->fp_code_heap;
+   else
    if (prog->type == PIPE_SHADER_GEOMETRY) heap = nv50->screen->gp_code_heap;
    else
       heap = nv50->screen->vp_code_heap;
@@ -145,14 +146,16 @@ nv50_program_validate(struct nv50_context *nv50, struct nv50_program *prog)
    size = align(prog->code_size, 0x100);
 
    ret = nouveau_resource_alloc(heap, size, prog, &prog->res);
-   if (ret)
+   if (ret) {
+      NOUVEAU_ERR("out of code space for shader type %i\n", prog->type);
       return FALSE;
+   }
    prog->code_base = prog->res->start;
 
    nv50_relocate_program(prog, prog->code_base, 0);
 
    nv50_sifc_linear_u8(&nv50->base, nv50->screen->code,
-                       (prog->type << 16) + prog->code_base,
+                       (prog->type << NV50_CODE_BO_SIZE_LOG2) + prog->code_base,
                        NOUVEAU_BO_VRAM, prog->code_size, prog->code);
 
    BEGIN_RING(nv50->screen->base.channel, RING_3D(CODE_CB_FLUSH), 1);