nv50,nvc0: fix depth/stencil resolve

Cannot sample depth/stencil with a single view, and needed to use
different shader code for nve4.

4 files changed, 65 insertions, 26 deletions

diff --git a/src/gallium/drivers/nv50/nv50_formats.c b/src/gallium/drivers/nv50/nv50_formats.c
index a009b27c861..41757d2a00d 100644
--- a/src/gallium/drivers/nv50/nv50_formats.c
+++ b/src/gallium/drivers/nv50/nv50_formats.c
@@ -60,6 +60,7 @@
 #endif
 
 #define NV50_SURFACE_FORMAT_NONE 0
+#define NV50_ZETA_FORMAT_NONE 0
 
 /* for vertex buffers: */
 #define NV50_TIC_0_FMT_8_8_8    NV50_TIC_0_FMT_8_8_8_8
@@ -167,10 +168,13 @@ const struct nv50_format nv50_format_table[PIPE_FORMAT_COUNT] =
    ZXB(Z16_UNORM, Z16_UNORM, C0, C0, C0, xx, UNORM, Z16, TZ),
    ZXB(Z32_FLOAT, Z32_FLOAT, C0, C0, C0, xx, FLOAT, Z32, TZ),
    ZXB(Z24X8_UNORM, Z24_X8_UNORM, C0, C0, C0, xx, UNORM, Z24_X8, TZ),
-   ZSB(Z24_UNORM_S8_UINT, Z24_S8_UNORM, C0, C1, C0, xx, UNORM, Z24_S8, TZ),
-   SZB(S8_UINT_Z24_UNORM, S8_Z24_UNORM, C1, C0, C1, xx, UNORM, S8_Z24, TZ),
-   ZSB(Z32_FLOAT_S8X24_UINT, Z32_S8_X24_FLOAT, C0, C1, C0, xx, FLOAT,
+   ZSB(Z24_UNORM_S8_UINT, Z24_S8_UNORM, C0, C0, C0, xx, UNORM, Z24_S8, TZ),
+   ZSB(X24S8_UINT, NONE, C1, C1, C1, xx, UNORM, Z24_S8, T),
+   SZB(S8_UINT_Z24_UNORM, S8_Z24_UNORM, C1, C1, C1, xx, UNORM, S8_Z24, TZ),
+   SZB(S8X24_UINT, NONE, C0, C0, C0, xx, UNORM, S8_Z24, T),
+   ZSB(Z32_FLOAT_S8X24_UINT, Z32_S8_X24_FLOAT, C0, C0, C0, xx, FLOAT,
        Z32_S8_X24, TZ),
+   ZSB(X32_S8X24_UINT, NONE, C1, C1, C1, xx, FLOAT, Z32_S8_X24, T),
 
    F3B(B5G6R5_UNORM, B5G6R5_UNORM, C2, C1, C0, xx, UNORM, 5_6_5, T),
    C4B(B5G5R5A1_UNORM, BGR5_A1_UNORM, C2, C1, C0, C3, UNORM, 5_5_5_1, TB),
diff --git a/src/gallium/drivers/nv50/nv50_resource.h b/src/gallium/drivers/nv50/nv50_resource.h
index 5d5596ea6e3..920fa0a4a2b 100644
--- a/src/gallium/drivers/nv50/nv50_resource.h
+++ b/src/gallium/drivers/nv50/nv50_resource.h
@@ -96,6 +96,18 @@ nv50_surface(struct pipe_surface *ps)
    return (struct nv50_surface *)ps;
 }
 
+static INLINE enum pipe_format
+nv50_zs_to_s_format(enum pipe_format format)
+{
+   switch (format) {
+   case PIPE_FORMAT_Z24_UNORM_S8_UINT: return PIPE_FORMAT_X24S8_UINT;
+   case PIPE_FORMAT_S8_UINT_Z24_UNORM: return PIPE_FORMAT_S8X24_UINT;
+   case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT: return PIPE_FORMAT_X32_S8X24_UINT;
+   default:
+      return format;
+   }
+}
+
 #ifndef __NVC0_RESOURCE_H__
 
 unsigned
diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c
index 0d29dcc8a2a..b38e49ffcc1 100644
--- a/src/gallium/drivers/nv50/nv50_surface.c
+++ b/src/gallium/drivers/nv50/nv50_surface.c
@@ -207,7 +207,8 @@ nv50_resource_copy_region(struct pipe_context *pipe,
       return;
    }
 
-   assert(src->nr_samples == dst->nr_samples);
+   /* 0 and 1 are equal, only supporting 0/1, 2, 4 and 8 */
+   assert((src->nr_samples | 1) == (dst->nr_samples | 1));
 
    m2mf = (src->format == dst->format) ||
       (util_format_get_blocksizebits(src->format) ==
@@ -433,8 +434,8 @@ struct nv50_blitctx
       struct nv50_program *fp;
       unsigned num_textures[3];
       unsigned num_samplers[3];
-      struct pipe_sampler_view *texture;
-      struct nv50_tsc_entry *sampler;
+      struct pipe_sampler_view *texture[2];
+      struct nv50_tsc_entry *sampler[2];
       unsigned dirty;
    } saved;
    struct nv50_program vp;
@@ -495,17 +496,21 @@ nv50_blitctx_make_fp(struct nv50_blitctx *blit)
       /* 3 coords ZS in, S encoded in R, Z encoded in GBA (8_UNORM) */
       0x80000000, /* interp $r0 v[0x00] */
       0x80010004, /* interp $r1 v[0x04] */
-      0x80020009, /* interp $r2 flat v[0x8] */
-      0x00040780,
-      0xf6800001, /* texauto live { $r0,1,#,# } $t0 $s0 { $r0,1,2 } */
+      0x80020108, /* interp $r2 flat v[0x8] */
+      0x10008010, /* mov b32 $r4 $r0 */
+      0xf2820201, /* texauto live { $r0,#,#,# } $t1 $s1 { $r0,1,2 } */
+      0x00000784,
+      0xa000000d, /* cvt f32 $r3 s32 $r0 */
+      0x44014780,
+      0x10000801, /* mov b32 $r0 $r4 */
+      0x0403c780,
+      0xf2800001, /* texauto live { $r0,#,#,# } $t0 $s0 { $r0,1,2 } */
       0x00000784,
       0xc03f0009, /* mul f32 $r2 $r0 (2^24 - 1) */
       0x04b7ffff,
-      0xa0000201, /* cvt f32 $r0 s32 $r1 */
-      0x44014780,
       0xa0000409, /* cvt rni s32 $r2 f32 $r2 */
       0x8c004780,
-      0xc0010001, /* mul f32 $r0 $r0 1/0xff */
+      0xc0010601, /* mul f32 $r0 $r3 1/0xff */
       0x03b8080b,
       0xd03f0405, /* and b32 $r1 $r2 0x0000ff */
       0x0000000f,
@@ -531,14 +536,18 @@ nv50_blitctx_make_fp(struct nv50_blitctx *blit)
       /* 3 coords ZS in, Z encoded in RGB, S encoded in A (U8_UNORM) */
       0x80000000, /* interp $r0 v[0x00] */
       0x80010004, /* interp $r1 v[0x04] */
-      0x80020009, /* interp $r2 flat v[0x8] */
-      0x00040780,
-      0xf6800001, /* texauto live { $r0,1,#,# } $t0 $s0 { $r0,1,2 } */
+      0x80020108, /* interp $r2 flat v[0x8] */
+      0x10008010, /* mov b32 $r4 $r0 */
+      0xf2820201, /* texauto live { $r0,#,#,# } $t1 $s1 { $r0,1,2 } */
+      0x00000784,
+      0xa000000d, /* cvt f32 $r3 s32 $r0 */
+      0x44014780,
+      0x10000801, /* mov b32 $r0 $r4 */
+      0x0403c780,
+      0xf2800001, /* texauto live { $r0,#,#,# } $t0 $s0 { $r0,1,2 } */
       0x00000784,
       0xc03f0009, /* mul f32 $r2 $r0 (2^24 - 1) */
       0x04b7ffff,
-      0xa0000281, /* cvt f32 $r3 s32 $r1 */
-      0x44014780,
       0xa0000409, /* cvt rni s32 $r2 f32 $r2 */
       0x8c004780,
       0xc001060d, /* mul f32 $r3 $r3 1/0xff */
@@ -569,7 +578,7 @@ nv50_blitctx_make_fp(struct nv50_blitctx *blit)
    blit->fp.translated = TRUE;
    blit->fp.code = (uint32_t *)code; /* const_cast */
    blit->fp.code_size = sizeof(code);
-   blit->fp.max_gpr = 4;
+   blit->fp.max_gpr = 5;
    blit->fp.max_out = 4;
    blit->fp.in_nr = 1;
    blit->fp.in[0].mask = 0x7; /* last component flat */
@@ -702,11 +711,19 @@ nv50_blit_set_src(struct nv50_context *nv50,
    templ.swizzle_a = PIPE_SWIZZLE_ALPHA;
 
    nv50->textures[2][0] = nv50_create_sampler_view(pipe, res, &templ);
+   nv50->textures[2][0] = NULL;
 
    nv50_blit_fixup_tic_entry(nv50->textures[2][0]);
 
    nv50->num_textures[0] = nv50->num_textures[1] = 0;
    nv50->num_textures[2] = 1;
+
+   templ.format = nv50_zs_to_s_format(res->format);
+   if (templ.format != res->format) {
+      nv50->textures[2][1] = nv50_create_sampler_view(pipe, res, &templ);
+      nv50_blit_fixup_tic_entry(nv50->textures[2][1]);
+      nv50->num_textures[2] = 2;
+   }
 }
 
 static void
@@ -781,13 +798,16 @@ nv50_blitctx_pre_blit(struct nv50_blitctx *blit, struct nv50_context *nv50)
       blit->saved.num_textures[s] = nv50->num_textures[s];
       blit->saved.num_samplers[s] = nv50->num_samplers[s];
    }
-   blit->saved.texture = nv50->textures[2][0];
-   blit->saved.sampler = nv50->samplers[2][0];
+   blit->saved.texture[0] = nv50->textures[2][0];
+   blit->saved.texture[1] = nv50->textures[2][1];
+   blit->saved.sampler[0] = nv50->samplers[2][0];
+   blit->saved.sampler[1] = nv50->samplers[2][1];
 
    nv50->samplers[2][0] = &blit->sampler[blit->filter];
+   nv50->samplers[2][1] = &blit->sampler[blit->filter];
 
    nv50->num_samplers[0] = nv50->num_samplers[1] = 0;
-   nv50->num_samplers[2] = 1;
+   nv50->num_samplers[2] = 2;
 
    blit->saved.dirty = nv50->dirty;
 
@@ -815,13 +835,16 @@ nv50_blitctx_post_blit(struct nv50_context *nv50, struct nv50_blitctx *blit)
    nv50->fragprog = blit->saved.fp;
 
    pipe_sampler_view_reference(&nv50->textures[2][0], NULL);
+   pipe_sampler_view_reference(&nv50->textures[2][1], NULL);
 
    for (s = 0; s < 3; ++s) {
       nv50->num_textures[s] = blit->saved.num_textures[s];
       nv50->num_samplers[s] = blit->saved.num_samplers[s];
    }
-   nv50->textures[2][0] = blit->saved.texture;
-   nv50->samplers[2][0] = blit->saved.sampler;
+   nv50->textures[2][0] = blit->saved.texture[0];
+   nv50->textures[2][1] = blit->saved.texture[1];
+   nv50->samplers[2][0] = blit->saved.sampler[0];
+   nv50->samplers[2][1] = blit->saved.sampler[1];
 
    nv50->dirty = blit->saved.dirty |
       (NV50_NEW_FRAMEBUFFER | NV50_NEW_SCISSOR | NV50_NEW_SAMPLE_MASK |
diff --git a/src/gallium/drivers/nv50/nv50_texture.xml.h b/src/gallium/drivers/nv50/nv50_texture.xml.h
index 2b140be8d80..34accd81741 100644
--- a/src/gallium/drivers/nv50/nv50_texture.xml.h
+++ b/src/gallium/drivers/nv50/nv50_texture.xml.h
@@ -8,7 +8,7 @@ http://0x04.net/cgit/index.cgi/rules-ng-ng
 git clone git://0x04.net/rules-ng-ng
 
 The rules-ng-ng source files this header was generated from are:
-- rnndb/nv50_texture.xml (   8111 bytes, from 2012-03-31 16:47:45)
+- rnndb/nv50_texture.xml (   8111 bytes, from 2012-04-29 11:17:45)
 - ./rnndb/copyright.xml  (   6452 bytes, from 2011-08-11 18:25:12)
 - ./rnndb/nvchipsets.xml (   3701 bytes, from 2012-03-22 20:40:59)
 - ./rnndb/nv50_defs.xml  (   5468 bytes, from 2011-08-11 18:25:12)
@@ -147,8 +147,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define NV50_TIC_0_FMT_RGTC1					0x00000027
 #define NV50_TIC_0_FMT_RGTC2					0x00000028
 #define NV50_TIC_0_FMT_S8_Z24					0x00000029
-#define NV50_TIC_0_FMT_Z24_S8					0x0000002a
-#define NV50_TIC_0_FMT_Z24_X8					0x0000002b
+#define NV50_TIC_0_FMT_Z24_X8					0x0000002a
+#define NV50_TIC_0_FMT_Z24_S8					0x0000002b
 #define NV50_TIC_0_FMT_Z24_C8_MS4_CS4				0x0000002c
 #define NV50_TIC_0_FMT_Z24_C8_MS8_CS8				0x0000002d
 #define NV50_TIC_0_FMT_Z24_C8_MS4_CS12				0x0000002e