r300g: do not make copies of constant buffers, emit them directly

5 files changed, 27 insertions, 35 deletions

diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 55cec881490..d2b8aa10284 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -247,8 +247,8 @@ struct r300_ztop_state {
 
 struct r300_constant_buffer {
     /* Buffer of constants */
-    uint32_t constants[256][4];
-    /* Total number of constants */
+    uint32_t *ptr;
+    /* Total number of vec4s */
     unsigned count;
 };
 
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 10d9e675ac0..349f3912d13 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -171,6 +171,7 @@ void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat
     struct r300_fragment_shader *fs = r300_fs(r300);
     struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
     unsigned count = fs->shader->externals_count * 4;
+    unsigned i, j;
     CS_LOCALS(r300);
 
     if (count == 0)
@@ -178,7 +179,9 @@ void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat
 
     BEGIN_CS(size);
     OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, count);
-    OUT_CS_TABLE(buf->constants, count);
+    for (i = 0; i < count; i++)
+        for (j = 0; j < 4; j++)
+            OUT_CS(pack_float24(buf->ptr[i*4+j]));
     END_CS;
 }
 
@@ -190,7 +193,6 @@ void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo
     unsigned count = fs->shader->rc_state_count;
     unsigned first = fs->shader->externals_count;
     unsigned end = constants->Count;
-    uint32_t cdata[4];
     unsigned j;
     CS_LOCALS(r300);
 
@@ -203,11 +205,9 @@ void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo
             const float *data =
                     get_rc_constant_state(r300, &constants->Constants[i]);
 
-            for (j = 0; j < 4; j++)
-                cdata[j] = pack_float24(data[j]);
-
             OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
-            OUT_CS_TABLE(cdata, 4);
+            for (j = 0; j < 4; j++)
+                OUT_CS(pack_float24(data[j]));
         }
     }
     END_CS;
@@ -234,7 +234,7 @@ void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat
     BEGIN_CS(size);
     OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST);
     OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count);
-    OUT_CS_TABLE(buf->constants, count);
+    OUT_CS_TABLE(buf->ptr, count);
     END_CS;
 }
 
@@ -926,7 +926,7 @@ void r300_emit_vs_constants(struct r300_context* r300,
                (r300->screen->caps.is_r500 ?
                R500_PVS_CONST_START : R300_PVS_CONST_START));
     OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4);
-    OUT_CS_TABLE(buf->constants, count * 4);
+    OUT_CS_TABLE(buf->ptr, count * 4);
     END_CS;
 }
 
diff --git a/src/gallium/drivers/r300/r300_screen_buffer.c b/src/gallium/drivers/r300/r300_screen_buffer.c
index 15d3f25c776..51d044af71e 100644
--- a/src/gallium/drivers/r300/r300_screen_buffer.c
+++ b/src/gallium/drivers/r300/r300_screen_buffer.c
@@ -136,6 +136,9 @@ static void r300_buffer_destroy(struct pipe_screen *screen,
     struct r300_screen *r300screen = r300_screen(screen);
     struct r300_buffer *rbuf = r300_buffer(buf);
 
+    if (rbuf->constant_buffer)
+        FREE(rbuf->constant_buffer);
+
     r300_winsys_buffer_destroy(r300screen, rbuf);
     FREE(rbuf);
 }
@@ -154,10 +157,8 @@ r300_buffer_transfer_map( struct pipe_context *pipe,
 
     if (rbuf->user_buffer)
         return (uint8_t *) rbuf->user_buffer + transfer->box.x;
-
-    if (rbuf->b.b.bind & PIPE_BIND_CONSTANT_BUFFER) {
-	goto just_map;
-    }
+    if (rbuf->constant_buffer)
+        return (uint8_t *) rbuf->constant_buffer + transfer->box.x;
 
     /* check if the mapping is to a range we already flushed */
     if (transfer->usage & PIPE_TRANSFER_DISCARD) {
@@ -182,7 +183,6 @@ r300_buffer_transfer_map( struct pipe_context *pipe,
 	}
     }
 
-just_map:
     map = rws->buffer_map(rws, rbuf->buf, r300->cs, transfer->usage);
 
     if (map == NULL)
@@ -208,9 +208,8 @@ static void r300_buffer_transfer_flush_region( struct pipe_context *pipe,
 
     if (rbuf->user_buffer)
 	return;
-
-    if (rbuf->b.b.bind & PIPE_BIND_CONSTANT_BUFFER)
-	return;
+    if (rbuf->constant_buffer)
+        return;
 
     /* mark the range as used */
     for(i = 0; i < rbuf->num_ranges; ++i) {
@@ -270,6 +269,12 @@ struct pipe_resource *r300_buffer_create(struct pipe_screen *screen,
     rbuf->b.b.screen = screen;
     rbuf->domain = R300_DOMAIN_GTT;
 
+    /* Alloc constant buffers in RAM. */
+    if (templ->bind & PIPE_BIND_CONSTANT_BUFFER) {
+        rbuf->constant_buffer = MALLOC(templ->width0);
+        return &rbuf->b.b;
+    }
+
     rbuf->buf =
         r300screen->rws->buffer_create(r300screen->rws,
                                        rbuf->b.b.width0, alignment,
diff --git a/src/gallium/drivers/r300/r300_screen_buffer.h b/src/gallium/drivers/r300/r300_screen_buffer.h
index 4f7c0ec87e5..fdd5d875d90 100644
--- a/src/gallium/drivers/r300/r300_screen_buffer.h
+++ b/src/gallium/drivers/r300/r300_screen_buffer.h
@@ -55,6 +55,7 @@ struct r300_buffer
     enum r300_buffer_domain domain;
 
     void *user_buffer;
+    void *constant_buffer;
     struct r300_buffer_range ranges[R300_BUFFER_MAX_RANGES];
     unsigned num_ranges;
 };
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 8214bcc2464..8d849cd2442 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -1738,8 +1738,7 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
     struct r300_constant_buffer *cbuf;
-    struct pipe_transfer *tr;
-    float *mapped;
+    uint32_t *mapped = r300_buffer(buf)->user_buffer;
     int max_size = 0, max_size_bytes = 0, clamped_size = 0;
 
     switch (shader) {
@@ -1762,8 +1761,7 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
     max_size_bytes = max_size * 4 * sizeof(float);
 
     if (buf == NULL || buf->width0 == 0 ||
-        (mapped = pipe_buffer_map(pipe, buf, PIPE_TRANSFER_READ, &tr)) == NULL)
-    {
+        (mapped = r300_buffer(buf)->constant_buffer) == NULL) {
         cbuf->count = 0;
         return;
     }
@@ -1781,17 +1779,7 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
 
         clamped_size = MIN2(buf->width0, max_size_bytes);
         cbuf->count = clamped_size / (4 * sizeof(float));
-
-        if (shader == PIPE_SHADER_FRAGMENT && !r300->screen->caps.is_r500) {
-            unsigned i,j;
-
-            /* Convert constants to float24. */
-            for (i = 0; i < cbuf->count; i++)
-                for (j = 0; j < 4; j++)
-                    cbuf->constants[i][j] = pack_float24(mapped[i*4+j]);
-        } else {
-            memcpy(cbuf->constants, mapped, clamped_size);
-        }
+        cbuf->ptr = mapped;
     }
 
     if (shader == PIPE_SHADER_VERTEX) {
@@ -1807,8 +1795,6 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
     } else if (shader == PIPE_SHADER_FRAGMENT) {
         r300->fs_constants.dirty = TRUE;
     }
-
-    pipe_buffer_unmap(pipe, buf, tr);
 }
 
 void r300_init_state_functions(struct r300_context* r300)