r300g: add texture compare modes

6 files changed, 148 insertions, 37 deletions

diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 6bd2766730f..232530b7dc4 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -91,6 +91,8 @@ struct r300_rs_block {
 };
 
 struct r300_sampler_state {
+    struct pipe_sampler_state state;
+
     uint32_t filter0;      /* R300_TX_FILTER0: 0x4400 */
     uint32_t filter1;      /* R300_TX_FILTER1: 0x4440 */
     uint32_t border_color; /* R300_TX_BORDER_COLOR: 0x45c0 */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index d7b6511d6d5..199ce3a945d 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -158,6 +158,13 @@ static const float * get_shader_constant(
                     vec[1] = 1.0 / tex->height0;
                     break;
 
+                /* Texture compare-fail value. */
+                /* XXX Since Gallium doesn't support GL_ARB_shadow_ambient,
+                 * this is always (0,0,0,0). */
+                case RC_STATE_SHADOW_AMBIENT:
+                    vec[3] = 0;
+                    break;
+
                 default:
                     debug_printf("r300: Implementation error: "
                         "Unknown RC_CONSTANT type %d\n", constant->u.State[0]);
@@ -1030,18 +1037,20 @@ validate:
 
     if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) {
         if (r300screen->caps->is_r500) {
-            r500_emit_fragment_program_code(r300, &r300->fs->code);
+            r500_emit_fragment_program_code(r300, &r300->fs->shader->code);
         } else {
-            r300_emit_fragment_program_code(r300, &r300->fs->code);
+            r300_emit_fragment_program_code(r300, &r300->fs->shader->code);
         }
         r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER;
     }
 
     if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER_CONSTANTS) {
         if (r300screen->caps->is_r500) {
-            r500_emit_fs_constant_buffer(r300, &r300->fs->code.constants);
+            r500_emit_fs_constant_buffer(r300,
+                                         &r300->fs->shader->code.constants);
         } else {
-            r300_emit_fs_constant_buffer(r300, &r300->fs->code.constants);
+            r300_emit_fs_constant_buffer(r300,
+                                         &r300->fs->shader->code.constants);
         }
         r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS;
     }
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index 79b01bb4dc2..4e1b61ca409 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -22,6 +22,9 @@
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
 #include "tgsi/tgsi_dump.h"
 
 #include "r300_context.h"
@@ -33,8 +36,8 @@
 #include "radeon_compiler.h"
 
 /* Convert info about FS input semantics to r300_shader_semantics. */
-static void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
-                                       struct r300_shader_semantics* fs_inputs)
+void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
+                                struct r300_shader_semantics* fs_inputs)
 {
     int i;
     unsigned index;
@@ -66,7 +69,6 @@ static void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
     }
 }
 
-
 static void find_output_registers(struct r300_fragment_program_compiler * compiler,
                                   struct r300_fragment_shader * fs)
 {
@@ -95,7 +97,7 @@ static void allocate_hardware_inputs(
     void * mydata)
 {
     struct r300_shader_semantics* inputs =
-        &((struct r300_fragment_shader*)c->UserData)->inputs;
+        (struct r300_shader_semantics*)c->UserData;
     int i, reg = 0;
 
     /* Allocate input registers. */
@@ -114,31 +116,45 @@ static void allocate_hardware_inputs(
     }
 }
 
-void r300_translate_fragment_shader(struct r300_context* r300,
-                                    struct r300_fragment_shader* fs)
+static void get_compare_state(
+    struct r300_context* r300,
+    struct r300_fragment_program_external_state* state,
+    unsigned shadow_samplers)
+{
+    memset(state, 0, sizeof(*state));
+
+    for (int i = 0; i < r300->sampler_count; i++) {
+        struct r300_sampler_state* s = r300->sampler_states[i];
+
+        if (s && s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+            /* XXX Gallium doesn't provide us with any information regarding
+             * this mode, so we are screwed. I'm setting 0 = LUMINANCE. */
+            state->unit[i].depth_texture_mode = 0;
+
+            /* Fortunately, no need to translate this. */
+            state->unit[i].texture_compare_func = s->state.compare_func;
+        }
+    }
+}
+
+static void r300_translate_fragment_shader(
+    struct r300_context* r300,
+    struct r300_fragment_shader_code* shader)
 {
+    struct r300_fragment_shader* fs = r300->fs;
     struct r300_fragment_program_compiler compiler;
     struct tgsi_to_rc ttr;
 
-    /* Initialize. */
-    r300_shader_read_fs_inputs(&fs->info, &fs->inputs);
-
     /* Setup the compiler. */
     memset(&compiler, 0, sizeof(compiler));
     rc_init(&compiler.Base);
     compiler.Base.Debug = DBG_ON(r300, DBG_FP);
 
-    compiler.code = &fs->code;
+    compiler.code = &shader->code;
+    compiler.state = shader->compare_state;
     compiler.is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
     compiler.AllocateHwInputs = &allocate_hardware_inputs;
-    compiler.UserData = fs;
-
-    /* XXX: Program compilation depends on texture compare modes,
-     * which are sampler state. Therefore, programs need to be recompiled
-     * depending on this state as in the classic Mesa driver.
-     *
-     * This is not yet handled correctly.
-     */
+    compiler.UserData = &fs->inputs;
 
     find_output_registers(&compiler, fs);
 
@@ -153,6 +169,8 @@ void r300_translate_fragment_shader(struct r300_context* r300,
 
     r300_tgsi_to_rc(&ttr, fs->state.tokens);
 
+    fs->shadow_samplers = compiler.Base.Program.ShadowSamplers;
+
     /* Invoke the compiler */
     r3xx_compile_fragment_program(&compiler);
     if (compiler.Base.Error) {
@@ -164,5 +182,51 @@ void r300_translate_fragment_shader(struct r300_context* r300,
 
     /* And, finally... */
     rc_destroy(&compiler.Base);
-    fs->translated = TRUE;
+}
+
+boolean r300_pick_fragment_shader(struct r300_context* r300)
+{
+    struct r300_fragment_shader* fs = r300->fs;
+    struct r300_fragment_program_external_state state;
+    struct r300_fragment_shader_code* ptr;
+
+    if (!fs->first) {
+        /* Build the fragment shader for the first time. */
+        fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code);
+
+        /* BTW shadow samplers will be known after the first translation,
+         * therefore we set ~0, which means it should look at all sampler
+         * states. This choice doesn't have any impact on the correctness. */
+        get_compare_state(r300, &fs->shader->compare_state, ~0);
+        r300_translate_fragment_shader(r300, fs->shader);
+        return TRUE;
+
+    } else if (fs->shadow_samplers) {
+        get_compare_state(r300, &state, fs->shadow_samplers);
+
+        /* Check if the currently-bound shader has been compiled
+         * with the texture-compare state we need. */
+        if (memcmp(&fs->shader->compare_state, &state, sizeof(state)) != 0) {
+            /* Search for the right shader. */
+            ptr = fs->first;
+            while (ptr) {
+                if (memcmp(&ptr->compare_state, &state, sizeof(state)) == 0) {
+                    fs->shader = ptr;
+                    return TRUE;
+                }
+                ptr = ptr->next;
+            }
+
+            /* Not found, gotta compile a new one. */
+            ptr = CALLOC_STRUCT(r300_fragment_shader_code);
+            ptr->next = fs->first;
+            fs->first = fs->shader = ptr;
+
+            ptr->compare_state = state;
+            r300_translate_fragment_shader(r300, ptr);
+            return TRUE;
+        }
+    }
+
+    return FALSE;
 }
diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h
index 630e2d0c8a5..40ce874353c 100644
--- a/src/gallium/drivers/r300/r300_fs.h
+++ b/src/gallium/drivers/r300/r300_fs.h
@@ -30,6 +30,13 @@
 #include "radeon_code.h"
 #include "r300_shader_semantics.h"
 
+struct r300_fragment_shader_code {
+    struct r300_fragment_program_external_state compare_state;
+    struct rX00_fragment_program_code code;
+
+    struct r300_fragment_shader_code* next;
+};
+
 struct r300_fragment_shader {
     /* Parent class */
     struct pipe_shader_state state;
@@ -37,21 +44,28 @@ struct r300_fragment_shader {
     struct tgsi_shader_info info;
     struct r300_shader_semantics inputs;
 
-    /* Has this shader been translated yet? */
-    boolean translated;
+    /* Bits 0-15: TRUE if it's a shadow sampler, FALSE otherwise. */
+    unsigned shadow_samplers;
 
-    /* Compiled code */
-    struct rX00_fragment_program_code code;
+    /* Currently-bound fragment shader. */
+    struct r300_fragment_shader_code* shader;
+
+    /* List of the same shaders compiled with different texture-compare
+     * states. */
+    struct r300_fragment_shader_code* first;
 };
 
+void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
+                                struct r300_shader_semantics* fs_inputs);
 
-void r300_translate_fragment_shader(struct r300_context* r300,
-                                    struct r300_fragment_shader* fs);
+/* Return TRUE if the shader was switched and should be re-emitted. */
+boolean r300_pick_fragment_shader(struct r300_context* r300);
 
-static inline boolean r300_fragment_shader_writes_depth(struct r300_fragment_shader *fs)
+static INLINE boolean r300_fragment_shader_writes_depth(struct r300_fragment_shader *fs)
 {
     if (!fs)
-	return FALSE;
-    return (fs->code.writes_depth) ? TRUE : FALSE;
+        return FALSE;
+    return (fs->shader->code.writes_depth) ? TRUE : FALSE;
 }
+
 #endif /* R300_FS_H */
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 5f332b2e0c2..8bcd6c50608 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -355,6 +355,7 @@ static void* r300_create_fs_state(struct pipe_context* pipe,
     fs->state.tokens = tgsi_dup_tokens(shader->tokens);
 
     tgsi_scan_shader(shader->tokens, &fs->info);
+    r300_shader_read_fs_inputs(&fs->info, &fs->inputs);
 
     return (void*)fs;
 }
@@ -368,11 +369,10 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
     if (fs == NULL) {
         r300->fs = NULL;
         return;
-    } else if (!fs->translated) {
-        r300_translate_fragment_shader(r300, fs);
     }
 
     r300->fs = fs;
+    r300_pick_fragment_shader(r300);
 
     r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
 }
@@ -381,7 +381,14 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
 static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
 {
     struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
-    rc_constants_destroy(&fs->code.constants);
+    struct r300_fragment_shader_code *tmp, *ptr = fs->first;
+
+    while (ptr) {
+        tmp = ptr;
+        ptr = ptr->next;
+        rc_constants_destroy(&tmp->code.constants);
+        FREE(tmp);
+    }
     FREE((void*)fs->state.tokens);
     FREE(shader);
 }
@@ -547,6 +554,8 @@ static void*
     int lod_bias;
     union util_color uc;
 
+    sampler->state = *state;
+
     sampler->filter0 |=
         (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
         (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) |
@@ -597,6 +606,14 @@ static void r300_bind_sampler_states(struct pipe_context* pipe,
     }
 
     r300->sampler_count = count;
+
+    /* Pick a fragment shader based on the texture compare state. */
+    if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) {
+        if (r300_pick_fragment_shader(r300)) {
+            r300->dirty_state |= R300_NEW_FRAGMENT_SHADER |
+                                 R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+        }
+    }
 }
 
 static void r300_lacks_vertex_textures(struct pipe_context* pipe,
diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index 9fb2de24032..096cdb20bbe 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -212,7 +212,8 @@ static void transform_srcreg(
     dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
 }
 
-static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src)
+static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
+                              uint32_t *shadowSamplers)
 {
     switch(src.Texture) {
         case TGSI_TEXTURE_1D:
@@ -233,14 +234,17 @@ static void transform_texture(struct rc_instruction * dst, struct tgsi_instructi
         case TGSI_TEXTURE_SHADOW1D:
             dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
             dst->U.I.TexShadow = 1;
+            *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
             break;
         case TGSI_TEXTURE_SHADOW2D:
             dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
             dst->U.I.TexShadow = 1;
+            *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
             break;
         case TGSI_TEXTURE_SHADOWRECT:
             dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
             dst->U.I.TexShadow = 1;
+            *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
             break;
     }
 }
@@ -269,7 +273,8 @@ static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_inst
 
     /* Texturing. */
     if (src->Instruction.Texture)
-       transform_texture(dst, src->Texture);
+        transform_texture(dst, src->Texture,
+                          &ttr->compiler->Program.ShadowSamplers);
 }
 
 static void handle_immediate(struct tgsi_to_rc * ttr, struct tgsi_full_immediate * imm)