1 files changed, 184 insertions, 54 deletions

diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index eeb97a2d37..199ce3a945 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -1,5 +1,6 @@
 /*
  * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2009 Marek Olšák <maraeo@gmail.com>
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -22,6 +23,7 @@
 
 /* r300_emit: Functions for emitting state. */
 
+#include "util/u_format.h"
 #include "util/u_math.h"
 
 #include "r300_context.h"
@@ -40,9 +42,16 @@ void r300_emit_blend_state(struct r300_context* r300,
     CS_LOCALS(r300);
     BEGIN_CS(8);
     OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3);
-    OUT_CS(blend->blend_control);
-    OUT_CS(blend->alpha_blend_control);
-    OUT_CS(blend->color_channel_mask);
+    if (r300->framebuffer_state.nr_cbufs) {
+        OUT_CS(blend->blend_control);
+        OUT_CS(blend->alpha_blend_control);
+        OUT_CS(blend->color_channel_mask);
+    } else {
+        OUT_CS(0);
+        OUT_CS(0);
+        OUT_CS(0);
+        /* XXX also disable fastfill here once it's supported */
+    }
     OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop);
     OUT_CS_REG(R300_RB3D_DITHER_CTL, blend->dither);
     END_CS;
@@ -129,7 +138,9 @@ static const float * get_shader_constant(
     struct rc_constant * constant,
     struct r300_constant_buffer * externals)
 {
-    static const float zero[4] = { 0.0, 0.0, 0.0, 0.0 };
+    static float vec[4] = { 0.0, 0.0, 0.0, 1.0 };
+    struct pipe_texture *tex;
+
     switch(constant->Type) {
         case RC_CONSTANT_EXTERNAL:
             return externals->constants[constant->u.External];
@@ -137,11 +148,38 @@ static const float * get_shader_constant(
         case RC_CONSTANT_IMMEDIATE:
             return constant->u.Immediate;
 
+        case RC_CONSTANT_STATE:
+            switch (constant->u.State[0]) {
+                /* Factor for converting rectangle coords to
+                 * normalized coords. Should only show up on non-r500. */
+                case RC_STATE_R300_TEXRECT_FACTOR:
+                    tex = &r300->textures[constant->u.State[1]]->tex;
+                    vec[0] = 1.0 / tex->width0;
+                    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]);
+            }
+            break;
+
         default:
-            debug_printf("r300: Implementation error: Unhandled constant type %i\n",
-                constant->Type);
-            return zero;
+            debug_printf("r300: Implementation error: "
+                "Unhandled constant type %d\n", constant->Type);
     }
+
+    /* This should either be (0, 0, 0, 1), which should be a relatively safe
+     * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
+     * state factors. */
+    return vec;
 }
 
 /* Convert a normal single-precision float into the 7.16 format
@@ -254,7 +292,7 @@ void r500_emit_fragment_program_code(struct r300_context* r300,
 
     BEGIN_CS(13 +
              ((code->inst_end + 1) * 6));
-    OUT_CS_REG(R500_US_CONFIG, 0);
+    OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
     OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx);
     OUT_CS_REG(R500_US_CODE_RANGE,
                R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
@@ -308,7 +346,13 @@ void r300_emit_fb_state(struct r300_context* r300,
     int i;
     CS_LOCALS(r300);
 
-    BEGIN_CS((10 * fb->nr_cbufs) + (fb->zsbuf ? 10 : 0) + 4);
+    /* Shouldn't fail unless there is a bug in the state tracker. */
+    assert(fb->nr_cbufs <= 4);
+
+    BEGIN_CS((10 * fb->nr_cbufs) + (2 * (4 - fb->nr_cbufs)) +
+             (fb->zsbuf ? 10 : 0) + 6);
+
+    /* Flush and free renderbuffer caches. */
     OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT,
         R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
         R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
@@ -316,6 +360,10 @@ void r300_emit_fb_state(struct r300_context* r300,
         R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
         R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
 
+    /* Set the number of colorbuffers. */
+    OUT_CS_REG(R300_RB3D_CCTL, R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs));
+
+    /* Set up colorbuffers. */
     for (i = 0; i < fb->nr_cbufs; i++) {
         surf = fb->cbufs[i];
         tex = (struct r300_texture*)surf->texture;
@@ -333,6 +381,12 @@ void r300_emit_fb_state(struct r300_context* r300,
             r300_translate_out_fmt(surf->format));
     }
 
+    /* Disable unused colorbuffers. */
+    for (; i < 4; i++) {
+        OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), R300_US_OUT_FMT_UNUSED);
+    }
+
+    /* Set up a zbuffer. */
     if (fb->zsbuf) {
         surf = fb->zsbuf;
         tex = (struct r300_texture*)surf->texture;
@@ -360,8 +414,6 @@ static void r300_emit_query_start(struct r300_context *r300)
     if (!query)
 	return;
 
-    /* XXX This will almost certainly not return good results
-     * for overlapping queries. */
     BEGIN_CS(4);
     if (caps->family == CHIP_FAMILY_RV530) {
         OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
@@ -543,24 +595,36 @@ void r300_emit_rs_block_state(struct r300_context* r300,
     END_CS;
 }
 
-void r300_emit_scissor_state(struct r300_context* r300,
-                             struct r300_scissor_state* scissor)
+static void r300_emit_scissor_regs(struct r300_context* r300,
+                                   struct r300_scissor_regs* scissor)
 {
     CS_LOCALS(r300);
 
     BEGIN_CS(3);
     OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
-    OUT_CS(scissor->scissor_top_left);
-    OUT_CS(scissor->scissor_bottom_right);
+    OUT_CS(scissor->top_left);
+    OUT_CS(scissor->bottom_right);
     END_CS;
 }
 
+void r300_emit_scissor_state(struct r300_context* r300,
+                             struct r300_scissor_state* scissor)
+{
+    if (r300->rs_state->rs.scissor) {
+        r300_emit_scissor_regs(r300, &scissor->scissor);
+    } else {
+        r300_emit_scissor_regs(r300, &scissor->framebuffer);
+    }
+}
+
 void r300_emit_texture(struct r300_context* r300,
                        struct r300_sampler_state* sampler,
                        struct r300_texture* tex,
                        unsigned offset)
 {
     uint32_t filter0 = sampler->filter0;
+    uint32_t format0 = tex->state.format0;
+    unsigned min_level, max_level;
     CS_LOCALS(r300);
 
     /* to emulate 1D textures through 2D ones correctly */
@@ -569,13 +633,20 @@ void r300_emit_texture(struct r300_context* r300,
         filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
     }
 
+    /* determine min/max levels */
+    /* the MAX_MIP level is the largest (finest) one */
+    max_level = MIN2(sampler->max_lod, tex->tex.last_level);
+    min_level = MIN2(sampler->min_lod, max_level);
+    format0 |= R300_TX_NUM_LEVELS(max_level);
+    filter0 |= R300_TX_MAX_MIP_LEVEL(min_level);
+
     BEGIN_CS(16);
     OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 |
         (offset << 28));
     OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1);
     OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color);
 
-    OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), tex->state.format0);
+    OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0);
     OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1);
     OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2);
     OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1);
@@ -584,50 +655,68 @@ void r300_emit_texture(struct r300_context* r300,
     END_CS;
 }
 
-/* XXX I can't read this and that's not good */
-void r300_emit_aos(struct r300_context* r300, unsigned offset)
+static boolean r300_validate_aos(struct r300_context *r300)
 {
     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
     struct pipe_vertex_element *velem = r300->vertex_element;
-    CS_LOCALS(r300);
     int i;
-    unsigned aos_count = r300->vertex_element_count;
 
+    /* Check if formats and strides are aligned to the size of DWORD. */
+    for (i = 0; i < r300->vertex_element_count; i++) {
+        if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
+            util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
+            return FALSE;
+        }
+    }
+    return TRUE;
+}
+
+void r300_emit_aos(struct r300_context* r300, unsigned offset)
+{
+    struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer;
+    struct pipe_vertex_element *velem = r300->vertex_element;
+    int i;
+    unsigned size1, size2, aos_count = r300->vertex_element_count;
     unsigned packet_size = (aos_count * 3 + 1) / 2;
+    CS_LOCALS(r300);
+
+    /* XXX Move this checking to a more approriate place. */
+    if (!r300_validate_aos(r300)) {
+        /* XXX We should fallback using Draw. */
+        assert(0);
+    }
+
     BEGIN_CS(2 + packet_size + aos_count * 2);
     OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
     OUT_CS(aos_count);
+
     for (i = 0; i < aos_count - 1; i += 2) {
-        int buf_num1 = velem[i].vertex_buffer_index;
-        int buf_num2 = velem[i+1].vertex_buffer_index;
-        assert(vbuf[buf_num1].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0);
-        assert(vbuf[buf_num2].stride % 4 == 0 && pf_get_size(velem[i+1].src_format) % 4 == 0);
-        OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num1].stride << 6) |
-               (pf_get_size(velem[i+1].src_format) << 14) | (vbuf[buf_num2].stride << 22));
-        OUT_CS(vbuf[buf_num1].buffer_offset + velem[i].src_offset +
-               offset * vbuf[buf_num1].stride);
-        OUT_CS(vbuf[buf_num2].buffer_offset + velem[i+1].src_offset +
-               offset * vbuf[buf_num2].stride);
+        vb1 = &vbuf[velem[i].vertex_buffer_index];
+        vb2 = &vbuf[velem[i+1].vertex_buffer_index];
+        size1 = util_format_get_blocksize(velem[i].src_format);
+        size2 = util_format_get_blocksize(velem[i+1].src_format);
+
+        OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
+               R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
+        OUT_CS(vb1->buffer_offset + velem[i].src_offset   + offset * vb1->stride);
+        OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
     }
+
     if (aos_count & 1) {
-        int buf_num = velem[i].vertex_buffer_index;
-        assert(vbuf[buf_num].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0);
-        OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num].stride << 6));
-        OUT_CS(vbuf[buf_num].buffer_offset + velem[i].src_offset +
-               offset * vbuf[buf_num].stride);
+        vb1 = &vbuf[velem[i].vertex_buffer_index];
+        size1 = util_format_get_blocksize(velem[i].src_format);
+
+        OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
+        OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
     }
 
-    /* XXX bare CS reloc */
     for (i = 0; i < aos_count; i++) {
-        cs_winsys->write_cs_reloc(cs_winsys,
-                                  vbuf[velem[i].vertex_buffer_index].buffer,
-                                  RADEON_GEM_DOMAIN_GTT,
-                                  0,
-                                  0);
-        cs_count -= 2;
+        OUT_CS_RELOC_NO_OFFSET(vbuf[velem[i].vertex_buffer_index].buffer,
+                               RADEON_GEM_DOMAIN_GTT, 0, 0);
     }
     END_CS;
 }
+
 #if 0
 void r300_emit_draw_packet(struct r300_context* r300)
 {
@@ -690,12 +779,22 @@ void r300_emit_vertex_format_state(struct r300_context* r300)
     END_CS;
 }
 
+
 void r300_emit_vertex_program_code(struct r300_context* r300,
                                    struct r300_vertex_program_code* code)
 {
     int i;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     unsigned instruction_count = code->length / 4;
+
+    int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72;
+    int input_count = MAX2(util_bitcount(code->InputsRead), 1);
+    int output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
+    int temp_count = MAX2(code->num_temporaries, 1);
+    int pvs_num_slots = MIN3(vtx_mem_size / input_count,
+                             vtx_mem_size / output_count, 10);
+    int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6);
+
     CS_LOCALS(r300);
 
     if (!r300screen->caps->has_tcl) {
@@ -708,8 +807,7 @@ void r300_emit_vertex_program_code(struct r300_context* r300,
     /* R300_VAP_PVS_CODE_CNTL_0
      * R300_VAP_PVS_CONST_CNTL
      * R300_VAP_PVS_CODE_CNTL_1
-     * See the r5xx docs for instructions on how to use these.
-     * XXX these could be optimized to select better values... */
+     * See the r5xx docs for instructions on how to use these. */
     OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3);
     OUT_CS(R300_PVS_FIRST_INST(0) |
             R300_PVS_XYZW_VALID_INST(instruction_count - 1) |
@@ -722,10 +820,11 @@ void r300_emit_vertex_program_code(struct r300_context* r300,
     for (i = 0; i < code->length; i++)
         OUT_CS(code->body.d[i]);
 
-    OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(10) |
-            R300_PVS_NUM_CNTLRS(5) |
+    OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) |
+            R300_PVS_NUM_CNTLRS(pvs_num_controllers) |
             R300_PVS_NUM_FPUS(r300screen->caps->num_vert_fpus) |
-            R300_PVS_VF_MAX_VTX_NUM(12));
+            R300_PVS_VF_MAX_VTX_NUM(12) |
+            (r300screen->caps->is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
     END_CS;
 }
 
@@ -790,13 +889,33 @@ void r300_emit_viewport_state(struct r300_context* r300,
     END_CS;
 }
 
+void r300_emit_texture_count(struct r300_context* r300)
+{
+    uint32_t tx_enable = 0;
+    int i;
+    CS_LOCALS(r300);
+
+    /* Notice that texture_count and sampler_count are just sizes
+     * of the respective arrays. We still have to check for the individual
+     * elements. */
+    for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) {
+        if (r300->textures[i]) {
+            tx_enable |= 1 << i;
+        }
+    }
+
+    BEGIN_CS(2);
+    OUT_CS_REG(R300_TX_ENABLE, tx_enable);
+    END_CS;
+
+}
+
 void r300_flush_textures(struct r300_context* r300)
 {
     CS_LOCALS(r300);
 
-    BEGIN_CS(4);
+    BEGIN_CS(2);
     OUT_CS_REG(R300_TX_INVALTAGS, 0);
-    OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1);
     END_CS;
 }
 
@@ -821,10 +940,17 @@ void r300_emit_dirty_state(struct r300_context* r300)
         return;
     }
 
+    /* Check size of CS. */
+    /* Make sure we have at least 8*1024 spare dwords. */
+    /* XXX It would be nice to know the number of dwords we really need to
+     * XXX emit. */
+    if (!r300->winsys->check_cs(r300->winsys, 8*1024)) {
+        r300->context.flush(&r300->context, 0, NULL);
+    }
+
     /* Clean out BOs. */
     r300->winsys->reset_bos(r300->winsys);
 
-    /* XXX check size */
 validate:
     /* Color buffers... */
     for (i = 0; i < r300->framebuffer_state.nr_cbufs; i++) {
@@ -911,18 +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;
     }
@@ -950,6 +1078,8 @@ validate:
     /* Samplers and textures are tracked separately but emitted together. */
     if (r300->dirty_state &
             (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) {
+        r300_emit_texture_count(r300);
+
         for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) {
   	    if (r300->dirty_state &
 		((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) {