Merge branch 'master' into r300-compiler

10 files changed, 85 insertions, 81 deletions

diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h
index 0a7e470363..883f0a02dc 100644
--- a/src/gallium/drivers/r300/r300_cs.h
+++ b/src/gallium/drivers/r300/r300_cs.h
@@ -68,11 +68,17 @@
 } while (0)
 
 #define OUT_CS(value) do { \
+    if (VERY_VERBOSE_CS || VERY_VERBOSE_REGISTERS) { \
+        DBG(cs_context_copy, DBG_CS, "r300: writing %08x\n", value); \
+    } \
     cs_winsys->write_cs_dword(cs_winsys, (value)); \
     cs_count--; \
 } while (0)
 
 #define OUT_CS_32F(value) do { \
+    if (VERY_VERBOSE_CS || VERY_VERBOSE_REGISTERS) { \
+        DBG(cs_context_copy, DBG_CS, "r300: writing %f\n", value); \
+    } \
     cs_winsys->write_cs_dword(cs_winsys, fui(value)); \
     cs_count--; \
 } while (0)
@@ -82,8 +88,9 @@
         DBG(cs_context_copy, DBG_CS, "r300: writing 0x%08X to register 0x%04X\n", \
             value, register); \
     assert(register); \
-    OUT_CS(CP_PACKET0(register, 0)); \
-    OUT_CS(value); \
+    cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0(register, 0)); \
+    cs_winsys->write_cs_dword(cs_winsys, value); \
+    cs_count -= 2; \
 } while (0)
 
 /* Note: This expects count to be the number of registers,
@@ -93,7 +100,8 @@
         DBG(cs_context_copy, DBG_CS, "r300: writing register sequence of %d to 0x%04X\n", \
             count, register); \
     assert(register); \
-    OUT_CS(CP_PACKET0(register, ((count) - 1))); \
+    cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0((register), ((count) - 1))); \
+    cs_count--; \
 } while (0)
 
 #define OUT_CS_RELOC(bo, offset, rd, wd, flags) do { \
@@ -101,9 +109,9 @@
             "domains (%d, %d, %d)\n", \
         bo, offset, rd, wd, flags); \
     assert(bo); \
-    OUT_CS(offset); \
+    cs_winsys->write_cs_dword(cs_winsys, offset); \
     cs_winsys->write_cs_reloc(cs_winsys, bo, rd, wd, flags); \
-    cs_count -= 2; \
+    cs_count -= 3; \
 } while (0)
 
 #define END_CS do { \
@@ -131,24 +139,26 @@
         DBG(cs_context_copy, DBG_CS, "r300: writing data sequence of %d to 0x%04X\n", \
             count, register); \
     assert(register); \
-    OUT_CS(CP_PACKET0(register, ((count) - 1)) | RADEON_ONE_REG_WR); \
+    cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0((register), ((count) - 1)) | RADEON_ONE_REG_WR); \
+    cs_count--; \
 } while (0)
 
 #define CP_PACKET3(op, count) \
     (RADEON_CP_PACKET3 | (op) | ((count) << 16))
 
 #define OUT_CS_PKT3(op, count) do { \
-    OUT_CS(CP_PACKET3(op, count)); \
+    cs_winsys->write_cs_dword(cs_winsys, CP_PACKET3(op, count)); \
+    cs_count--; \
 } while (0)
 
 #define OUT_CS_INDEX_RELOC(bo, offset, count, rd, wd, flags) do { \
     DBG(cs_context_copy, DBG_CS, "r300: writing relocation for index buffer %p," \
             "offset %d\n", bo, offset); \
     assert(bo); \
-    OUT_CS(offset); \
-    OUT_CS(count); \
+    cs_winsys->write_cs_dword(cs_winsys, offset); \
+    cs_winsys->write_cs_dword(cs_winsys, count); \
     cs_winsys->write_cs_reloc(cs_winsys, bo, rd, wd, flags); \
-    cs_count -= 2; \
+    cs_count -= 4; \
 } while (0)
 
 #endif /* R300_CS_H */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index a1b36ba2ed..77ce431cdc 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -381,6 +381,7 @@ void r300_emit_query_end(struct r300_context* r300,
             OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
             OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 0),
                     0, RADEON_GEM_DOMAIN_GTT, 0);
+	    break;
         default:
             debug_printf("r300: Implementation error: Chipset reports %d"
                     " pixel pipes!\n", caps->num_frag_pipes);
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index a0e848a59a..546ad545a5 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -126,9 +126,14 @@ void r300_translate_fragment_shader(struct r300_context* r300,
     /* Invoke the compiler */
     r3xx_compile_fragment_program(&compiler);
     if (compiler.Base.Error) {
-        /* Todo: Fail gracefully */
-        fprintf(stderr, "r300 FP: Compiler error\n");
-        abort();
+        /* Todo: Fallback to software rendering gracefully? */
+        fprintf(stderr, "r300 FP: Compiler error: %s\n", compiler.Base.ErrorMsg);
+
+        if (compiler.is_r500) {
+            memcpy(compiler.code, &r5xx_passthrough_fragment_shader, sizeof(r5xx_passthrough_fragment_shader));
+        } else {
+            memcpy(compiler.code, &r3xx_passthrough_fragment_shader, sizeof(r3xx_passthrough_fragment_shader));
+        }
     }
 
     /* And, finally... */
diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h
index 9fab789402..967e9f697e 100644
--- a/src/gallium/drivers/r300/r300_fs.h
+++ b/src/gallium/drivers/r300/r300_fs.h
@@ -48,4 +48,4 @@ struct r300_fragment_shader {
 void r300_translate_fragment_shader(struct r300_context* r300,
                                     struct r300_fragment_shader* fs);
 
-    #endif /* R300_FS_H */
+#endif /* R300_FS_H */
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 03cd219cde..3abff5db62 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -1478,6 +1478,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_TX_PITCH_EN                  (1 << 31)
 #       define R300_TX_WIDTH(x)                  ((x) << 0)
 #       define R300_TX_HEIGHT(x)                 ((x) << 11)
+#       define R300_TX_DEPTH(x)                  ((x) << 22)
 #       define R300_TX_NUM_LEVELS(x)             ((x) << 26)
 
 #define R300_TX_FORMAT1_0                   0x44C0
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index 737396d8d9..ca44e0f661 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -79,12 +79,14 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render,
     struct pipe_screen* screen = r300->context.screen;
     size_t size = (size_t)vertex_size * (size_t)count;
 
-    if (size + r300render->vbo_offset > r300render->vbo_size) 
+    if (size + r300render->vbo_offset > r300render->vbo_size)
     {
+        pipe_buffer_reference(&r300->vbo, NULL);
         r300render->vbo = pipe_buffer_create(screen,
                                              64,
                                              PIPE_BUFFER_USAGE_VERTEX,
                                              R300_MAX_VBO_SIZE);
+        r300render->vbo_offset = 0;
         r300render->vbo_size = R300_MAX_VBO_SIZE;
     }
 
@@ -117,7 +119,7 @@ static void r300_render_unmap_vertices(struct vbuf_render* render,
     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
     END_CS;
 
-    r300render->vbo_max_used = MAX2(r300render->vbo_max_used, 
+    r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
                                     r300render->vertex_size * (max + 1));
     pipe_buffer_unmap(screen, r300render->vbo);
 }
@@ -129,7 +131,6 @@ static void r300_render_release_vertices(struct vbuf_render* render)
 
     r300render->vbo_offset += r300render->vbo_max_used;
     r300render->vbo_max_used = 0;
-    r300->vbo = NULL;
 }
 
 static boolean r300_render_set_primitive(struct vbuf_render* render,
@@ -222,13 +223,6 @@ static void r300_render_draw(struct vbuf_render* render,
 
     r300_prepare_render(r300render, count);
 
-    /* Send our indices into an index buffer. */
-    index_buffer = pipe_buffer_create(screen, 64, PIPE_BUFFER_USAGE_VERTEX,
-                                      count * 2);
-    if (!index_buffer) {
-        return;
-    }
-
     BEGIN_CS(2 + (count+1)/2);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 3b5b1bbd37..f2659ca61f 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -101,11 +101,9 @@ static int r300_get_param(struct pipe_screen* pscreen, int param)
         case PIPE_CAP_MAX_RENDER_TARGETS:
             return 4;
         case PIPE_CAP_OCCLUSION_QUERY:
-            /* IN THEORY */
-            return 0;
+            return 1;
         case PIPE_CAP_TEXTURE_SHADOW_MAP:
-            /* IN THEORY */
-            return 0;
+            return 1;
         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
             if (r300screen->caps->is_r500) {
                 /* 13 == 4096x4096 */
@@ -347,16 +345,9 @@ static void* r300_transfer_map(struct pipe_screen* screen,
 {
     struct r300_texture* tex = (struct r300_texture*)transfer->texture;
     char* map;
-    unsigned flags = 0;
-
-    if (transfer->usage != PIPE_TRANSFER_WRITE) {
-        flags |= PIPE_BUFFER_USAGE_CPU_READ;
-    }
-    if (transfer->usage != PIPE_TRANSFER_READ) {
-        flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
-    }
 
-    map = pipe_buffer_map(screen, tex->buffer, flags);
+    map = pipe_buffer_map(screen, tex->buffer,
+                          pipe_transfer_buffer_flags(transfer));
 
     if (!map) {
         return NULL;
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 5f6b225d34..02b7ab9107 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -53,7 +53,7 @@ static void r300_vs_tab_routes(struct r300_context* r300,
     if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte)
     {
         for (i = 0; i < info->num_inputs; i++) {
-            switch (info->input_semantic_name[i]) {
+            switch (r300->vs->code.inputs[i]) {
                 case TGSI_SEMANTIC_POSITION:
                     pos = TRUE;
                     tab[i] = 0;
@@ -63,10 +63,12 @@ static void r300_vs_tab_routes(struct r300_context* r300,
                     cols++;
                     break;
                 case TGSI_SEMANTIC_PSIZE:
+                    assert(psize == FALSE);
                     psize = TRUE;
                     tab[i] = 15;
                     break;
                 case TGSI_SEMANTIC_FOG:
+                    assert(fog == FALSE);
                     fog = TRUE;
                     /* Fall through */
                 case TGSI_SEMANTIC_GENERIC:
@@ -125,7 +127,9 @@ static void r300_vs_tab_routes(struct r300_context* r300,
 
     vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
 
-    if (!pos) {
+    /* We need to add vertex position attribute only for SW TCL case,
+     * for HW TCL case it could be generated by vertex shader */
+    if (!pos && !r300screen->caps->has_tcl) {
         debug_printf("r300: Forcing vertex position attribute emit...\n");
         /* Make room for the position attribute
          * at the beginning of the tab. */
@@ -164,7 +168,7 @@ static void r300_vs_tab_routes(struct r300_context* r300,
         vinfo->hwfmt[3] |= (4 << (3 * i));
     }
 
-    for (i; i < texs; i++) {
+    for (; i < texs; i++) {
         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
@@ -334,48 +338,37 @@ static void r300_update_rs_block(struct r300_context* r300)
     struct r300_rs_block* rs = r300->rs_block;
     struct tgsi_shader_info* info = &r300->fs->info;
     int* tab = r300->vertex_info.fs_tab;
-    int col_count = 0, fp_offset = 0, i, memory_pos, tex_count = 0;
-
+    int col_count = 0, fp_offset = 0, i, tex_count = 0;
+    int rs_tex_comp = 0;
     memset(rs, 0, sizeof(struct r300_rs_block));
 
     if (r300_screen(r300->context.screen)->caps->is_r500) {
         for (i = 0; i < info->num_inputs; i++) {
             assert(tab[i] != -1);
-            memory_pos = tab[i] * 4;
             switch (info->input_semantic_name[i]) {
                 case TGSI_SEMANTIC_COLOR:
                     rs->ip[col_count] |=
-                        R500_RS_COL_PTR(memory_pos) |
+                        R500_RS_COL_PTR(col_count) |
                         R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
                     col_count++;
                     break;
                 case TGSI_SEMANTIC_GENERIC:
                     rs->ip[tex_count] |=
-                        R500_RS_SEL_S(memory_pos) |
-                        R500_RS_SEL_T(memory_pos + 1) |
-                        R500_RS_SEL_R(memory_pos + 2) |
-                        R500_RS_SEL_Q(memory_pos + 3);
+                        R500_RS_SEL_S(rs_tex_comp) |
+                        R500_RS_SEL_T(rs_tex_comp + 1) |
+                        R500_RS_SEL_R(rs_tex_comp + 2) |
+                        R500_RS_SEL_Q(rs_tex_comp + 3);
                     tex_count++;
+                    rs_tex_comp += 4;
                     break;
                 default:
                     break;
             }
         }
 
-        if (col_count == 0) {
-            rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
-        }
-
-        if (tex_count == 0) {
-            rs->ip[0] |=
-                R500_RS_SEL_S(R500_RS_IP_PTR_K0) |
-                R500_RS_SEL_T(R500_RS_IP_PTR_K0) |
-                R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
-                R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
-        }
-
         /* Rasterize at least one color, or bad things happen. */
         if ((col_count == 0) && (tex_count == 0)) {
+            rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
             col_count++;
         }
 
@@ -393,22 +386,22 @@ static void r300_update_rs_block(struct r300_context* r300)
     } else {
         for (i = 0; i < info->num_inputs; i++) {
             assert(tab[i] != -1);
-            memory_pos = tab[i] * 4;
             switch (info->input_semantic_name[i]) {
                 case TGSI_SEMANTIC_COLOR:
                     rs->ip[col_count] |=
-                        R300_RS_COL_PTR(memory_pos) |
+                        R300_RS_COL_PTR(col_count) |
                         R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
                     col_count++;
                     break;
                 case TGSI_SEMANTIC_GENERIC:
                     rs->ip[tex_count] |=
-                        R300_RS_TEX_PTR(memory_pos) |
+                        R300_RS_TEX_PTR(rs_tex_comp) |
                         R300_RS_SEL_S(R300_RS_SEL_C0) |
                         R300_RS_SEL_T(R300_RS_SEL_C1) |
                         R300_RS_SEL_R(R300_RS_SEL_C2) |
                         R300_RS_SEL_Q(R300_RS_SEL_C3);
                     tex_count++;
+                    rs_tex_comp+=4;
                     break;
                 default:
                     break;
@@ -445,7 +438,7 @@ static void r300_update_rs_block(struct r300_context* r300)
         }
     }
 
-    rs->count = (tex_count * 4) | (col_count << R300_IC_COUNT_SHIFT) |
+    rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) |
         R300_HIRES_EN;
 
     rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index 6e8c368320..ce60ded7ca 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -22,34 +22,39 @@
 
 #include "r300_texture.h"
 
-static void r300_setup_texture_state(struct r300_texture* tex,
-                                     unsigned width,
-                                     unsigned height,
-                                     unsigned levels)
+static void r300_setup_texture_state(struct r300_texture* tex)
 {
     struct r300_texture_state* state = &tex->state;
+    struct pipe_texture *pt = &tex->tex;
 
-    state->format0 = R300_TX_WIDTH((width - 1) & 0x7ff) |
-        R300_TX_HEIGHT((height - 1) & 0x7ff) |
-        R300_TX_NUM_LEVELS(levels) |
+    state->format0 = R300_TX_WIDTH((pt->width[0] - 1) & 0x7ff) |
+        R300_TX_HEIGHT((pt->height[0] - 1) & 0x7ff) |
+        R300_TX_DEPTH(util_logbase2(pt->depth[0]) & 0xf) |
+        R300_TX_NUM_LEVELS(pt->last_level) |
         R300_TX_PITCH_EN;
 
     /* XXX */
-    state->format1 = r300_translate_texformat(tex->tex.format);
+    state->format1 = r300_translate_texformat(pt->format);
+    if (pt->target == PIPE_TEXTURE_CUBE) {
+	state->format1 |= R300_TX_FORMAT_CUBIC_MAP;
+    }
+    if (pt->target == PIPE_TEXTURE_3D) {
+	state->format1 |= R300_TX_FORMAT_3D;
+    }
 
-    state->format2 = r300_texture_get_stride(tex, 0);
+    state->format2 = (r300_texture_get_stride(tex, 0) / pt->block.size) - 1;
 
     /* Assume (somewhat foolishly) that oversized textures will
      * not be permitted by the state tracker. */
-    if (width > 2048) {
+    if (pt->width[0] > 2048) {
         state->format2 |= R500_TXWIDTH_BIT11;
     }
-    if (height > 2048) {
+    if (pt->height[0] > 2048) {
         state->format2 |= R500_TXHEIGHT_BIT11;
     }
 
-    debug_printf("r300: Set texture state (%dx%d, pitch %d, %d levels)\n",
-            width, height, levels);
+    debug_printf("r300: Set texture state (%dx%d, %d levels)\n",
+		 pt->width[0], pt->height[0], pt->last_level);
 }
 
 /**
@@ -62,7 +67,7 @@ unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level)
         return tex->stride_override;
 
     if (level > tex->tex.last_level) {
-        debug_printf("%s: level (%u) > last_level (%u)\n", level, tex->tex.last_level);
+        debug_printf("%s: level (%u) > last_level (%u)\n", __FUNCTION__, level, tex->tex.last_level);
         return 0;
     }
 
@@ -120,8 +125,7 @@ static struct pipe_texture*
 
     r300_setup_miptree(tex);
 
-    r300_setup_texture_state(tex, template->width[0], template->height[0],
-                             template->last_level);
+    r300_setup_texture_state(tex);
 
     tex->buffer = screen->buffer_create(screen, 1024,
                                         PIPE_BUFFER_USAGE_PIXEL,
@@ -204,7 +208,7 @@ static struct pipe_texture*
     tex->stride_override = *stride;
 
     /* XXX */
-    r300_setup_texture_state(tex, tex->tex.width[0], tex->tex.height[0], 0);
+    r300_setup_texture_state(tex);
 
     pipe_buffer_reference(&tex->buffer, buffer);
 
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index 3109af5bac..78ee0f1611 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -72,6 +72,11 @@ static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
         /* W24_FP */
         case PIPE_FORMAT_Z24S8_UNORM:
             return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP);
+	/* Z5_Y6_X5 */
+        case PIPE_FORMAT_R16_SNORM:
+            return R300_EASY_TX_FORMAT(X, X, X, X, Z5Y6X5);
+        case PIPE_FORMAT_Z16_UNORM:
+	    return R300_EASY_TX_FORMAT(X, X, X, X, X16);
         default:
             debug_printf("r300: Implementation error: "
                 "Got unsupported texture format %s in %s\n",