preparation for r200 hw vertex programs. Increase R200_CMD_BUF_SZ to 16k instead of 8k (the reasons why it was exactly 8k are unclear). Add register defines, sanity code, fix potential issue with wrong statechange order when disabling fragment programs.

8 files changed, 91 insertions, 4 deletions

diff --git a/src/mesa/drivers/dri/r200/r200_context.h b/src/mesa/drivers/dri/r200/r200_context.h
index f6709d3d7fd..0de131adc1d 100644
--- a/src/mesa/drivers/dri/r200/r200_context.h
+++ b/src/mesa/drivers/dri/r200/r200_context.h
@@ -637,7 +637,7 @@ struct r200_dri_mirror {
 };
 
 
-#define R200_CMD_BUF_SZ  (8*1024) 
+#define R200_CMD_BUF_SZ  (16*1024) 
 
 struct r200_store {
    GLuint statenr;
diff --git a/src/mesa/drivers/dri/r200/r200_reg.h b/src/mesa/drivers/dri/r200/r200_reg.h
index 6d1b84769a8..dab08a9eb73 100644
--- a/src/mesa/drivers/dri/r200/r200_reg.h
+++ b/src/mesa/drivers/dri/r200/r200_reg.h
@@ -337,6 +337,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 /* gap */
 #define R200_SE_VAP_CNTL                           0x2080
 #define     R200_VAP_TCL_ENABLE                       0x00000001
+#define     R200_VAP_PROG_VTX_SHADER_ENABLE           0x00000004
 #define     R200_VAP_SINGLE_BUF_STATE_ENABLE          0x00000010
 #define     R200_VAP_FORCE_W_TO_ONE                   0x00010000
 #define     R200_VAP_D3D_TEX_DEFAULT                  0x00020000
@@ -660,6 +661,17 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R200_SE_TCL_POINT_SPRITE_CNTL     0x22c4
 #define     R200_POINTSIZE_SEL_STATE            (1<<16)
 /* gap */
+/* taken from r300, see comments there */
+#define R200_VAP_PVS_CNTL_1                 0x22d0
+#       define R200_PVS_CNTL_1_PROGRAM_START_SHIFT   0
+#       define R200_PVS_CNTL_1_POS_END_SHIFT         10
+#       define R200_PVS_CNTL_1_PROGRAM_END_SHIFT     20
+/* Addresses are relative the the vertex program parameters area. */
+#define R200_VAP_PVS_CNTL_2                 0x22d4
+#       define R200_PVS_CNTL_2_PARAM_OFFSET_SHIFT 0
+#       define R200_PVS_CNTL_2_PARAM_COUNT_SHIFT  16
+/* gap */
+
 #define R200_SE_VTX_ST_POS_0_X_4                   0x2300
 #define R200_SE_VTX_ST_POS_0_Y_4                   0x2304
 #define R200_SE_VTX_ST_POS_0_Z_4                   0x2308
@@ -1473,6 +1485,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R200_VS_PNT_SPRITE_CLAMP            0x000000BE
 #define R200_VS_MAX                         0x000001C0
 
+#define R200_PVS_PROG0                      0x00000080
+#define R200_PVS_PROG1                      0x00000180
+#define R200_PVS_PARAM0                     0x00000000
+#define R200_PVS_PARAM1                     0x00000100
 
 /*
  * Offsets in TCL scalar state
diff --git a/src/mesa/drivers/dri/r200/r200_sanity.c b/src/mesa/drivers/dri/r200/r200_sanity.c
index 4dc87cd945d..8c9319fe278 100644
--- a/src/mesa/drivers/dri/r200/r200_sanity.c
+++ b/src/mesa/drivers/dri/r200/r200_sanity.c
@@ -160,6 +160,7 @@ static struct {
    { R200_PP_TXFILTER_3, 8, "R200_PP_TXCTLALL_3"},
    { R200_PP_TXFILTER_4, 8, "R200_PP_TXCTLALL_4"},
    { R200_PP_TXFILTER_5, 8, "R200_PP_TXCTLALL_5"},
+   { R200_VAP_PVS_CNTL_1, 2, "R200_VAP_PVS_CNTL"},
 };
 
 struct reg_names {
@@ -532,6 +533,8 @@ static struct reg_names reg_names[] = {
    { R200_PP_TXCBLEND2_15, "R200_PP_TXCBLEND2_15" },
    { R200_PP_TXABLEND_15, "R200_PP_TXABLEND_15" },
    { R200_PP_TXABLEND2_15, "R200_PP_TXABLEND2_15" },
+   { R200_VAP_PVS_CNTL_1, "R200_VAP_PVS_CNTL_1" },
+   { R200_VAP_PVS_CNTL_2, "R200_VAP_PVS_CNTL_2" },
 };
 
 static struct reg_names scalar_names[] = {
@@ -935,6 +938,62 @@ static int radeon_emit_vectors(
    return 0;
 }
 
+static int radeon_emit_veclinear( 
+   drm_radeon_cmd_header_t header,
+   drm_radeon_cmd_buffer_t *cmdbuf )
+{
+   int sz = header.veclinear.count * 4;
+   int *data = (int *)cmdbuf->buf;
+   float *fdata =(float *)cmdbuf->buf;
+   int start = header.veclinear.addr_lo | (header.veclinear.addr_hi << 8);
+   int i;
+
+   if (1||VERBOSE)
+      fprintf(stderr, "emit vectors linear, start %d nr %d (end %d) (0x%x)\n",
+	      start, sz >> 2, start + (sz >> 2), header.i);
+
+
+   if (start < 0x60) {
+      for (i = 0 ; i < sz ;  i += 4) {
+	 fprintf(stderr, "R200_VS_PARAM %d 0 %f\n", (i >> 2) + start , fdata[i]);
+	 fprintf(stderr, "R200_VS_PARAM %d 1 %f\n", (i >> 2) + start, fdata[i+1]);
+	 fprintf(stderr, "R200_VS_PARAM %d 2 %f\n", (i >> 2) + start, fdata[i+2]);
+	 fprintf(stderr, "R200_VS_PARAM %d 3 %f\n", (i >> 2) + start, fdata[i+3]);
+      }
+   }
+   else if ((start >= 0x100) && (start < 0x160)) {
+      for (i = 0 ; i < sz ;  i += 4) {
+	 fprintf(stderr, "R200_VS_PARAM %d 0 %f\n", (i >> 2) + start - 0x100 + 0x60, fdata[i]);
+	 fprintf(stderr, "R200_VS_PARAM %d 1 %f\n", (i >> 2) + start - 0x100 + 0x60, fdata[i+1]);
+	 fprintf(stderr, "R200_VS_PARAM %d 2 %f\n", (i >> 2) + start - 0x100 + 0x60, fdata[i+2]);
+	 fprintf(stderr, "R200_VS_PARAM %d 3 %f\n", (i >> 2) + start - 0x100 + 0x60, fdata[i+3]);
+      }
+   }
+   else if ((start >= 0x80) && (start < 0xc0)) {
+      for (i = 0 ; i < sz ;  i += 4) {
+	 fprintf(stderr, "R200_VS_PROG %d OPDST %08x\n", (i >> 2) + start - 0x80, data[i]);
+	 fprintf(stderr, "R200_VS_PROG %d SRC1  %08x\n", (i >> 2) + start - 0x80, data[i+1]);
+	 fprintf(stderr, "R200_VS_PROG %d SRC2  %08x\n", (i >> 2) + start - 0x80, data[i+2]);
+	 fprintf(stderr, "R200_VS_PROG %d SRC3  %08x\n", (i >> 2)  + start - 0x80, data[i+3]);
+      }
+   }
+   else if ((start >= 0x180) && (start < 0x1c0)) {
+      for (i = start ; (i < start + sz) ;  i += 4) {
+	 fprintf(stderr, "R200_VS_PROG %d OPDST %08x\n", (i >> 2) + start - 0x180 + 0x40, data[i]);
+	 fprintf(stderr, "R200_VS_PROG %d SRC1  %08x\n", (i >> 2) + start - 0x180 + 0x40, data[i+1]);
+	 fprintf(stderr, "R200_VS_PROG %d SRC2  %08x\n", (i >> 2) + start - 0x180 + 0x40, data[i+2]);
+	 fprintf(stderr, "R200_VS_PROG %d SRC3  %08x\n", (i >> 2) + start - 0x180 + 0x40, data[i+3]);
+      }
+   }
+   else {
+      fprintf(stderr, "write to unknown vector area\n");
+   }
+
+   cmdbuf->buf += sz * sizeof(int);
+   cmdbuf->bufsz -= sz * sizeof(int);
+   return 0;
+}
+
 #if 0
 static int print_vertex_format( int vfmt )
 {
@@ -1360,6 +1419,13 @@ int r200SanityCmdBuffer( r200ContextPtr rmesa,
       case RADEON_CMD_WAIT:
 	 break;
 
+      case RADEON_CMD_VECLINEAR:
+	 if (radeon_emit_veclinear( header, &cmdbuf )) {
+	    fprintf(stderr,"radeon_emit_veclinear failed\n");
+	    return -EINVAL;
+	 }
+	 break;
+
       default:
 	 fprintf(stderr,"bad cmd_type %d at %p\n", 
 		   header.header.cmd_type,
diff --git a/src/mesa/drivers/dri/r200/r200_state.c b/src/mesa/drivers/dri/r200/r200_state.c
index 6ffb48c1505..41c7607b61d 100644
--- a/src/mesa/drivers/dri/r200/r200_state.c
+++ b/src/mesa/drivers/dri/r200/r200_state.c
@@ -2110,18 +2110,18 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
 	    if they didn't change) and restore tex coord routing */
 	 GLuint unit;
 	 for (unit = 0; unit < R200_MAX_TEXTURE_UNITS; unit++) {
+	    R200_STATECHANGE( rmesa, pix[unit] );
+	    R200_STATECHANGE( rmesa, tex[unit] );
 	    rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] &=
 		~(R200_TXFORMAT_ST_ROUTE_MASK | R200_TXFORMAT_LOOKUP_DISABLE);
 	    rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] |= unit << R200_TXFORMAT_ST_ROUTE_SHIFT;
 	    /* need to guard this with drmSupportsFragmentShader? Should never get here if
 	       we don't announce ATI_fs, right? */
 	    rmesa->hw.tex[unit].cmd[TEX_PP_TXMULTI_CTL] = 0;
-	    R200_STATECHANGE( rmesa, pix[unit] );
-	    R200_STATECHANGE( rmesa, tex[unit] );
          }
-	 rmesa->hw.cst.cmd[CST_PP_CNTL_X] = 0;
 	 R200_STATECHANGE( rmesa, cst );
 	 R200_STATECHANGE( rmesa, tf );
+	 rmesa->hw.cst.cmd[CST_PP_CNTL_X] = 0;
       }
       else {
 	 /* need to mark this dirty as pix/tf atoms have overwritten the data
diff --git a/src/mesa/drivers/dri/radeon/radeon_sanity.c b/src/mesa/drivers/dri/radeon/radeon_sanity.c
index 9ca9ebe84e0..557057784c3 100644
--- a/src/mesa/drivers/dri/radeon/radeon_sanity.c
+++ b/src/mesa/drivers/dri/radeon/radeon_sanity.c
@@ -156,6 +156,7 @@ static struct {
    { 0, 8, "R200_PP_TXCTLALL_3"},
    { 0, 8, "R200_PP_TXCTLALL_4"},
    { 0, 8, "R200_PP_TXCTLALL_5"},
+   { 0, 2, "R200_VAP_PVS_CNTL"},
 };
 
 struct reg_names {
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.c b/src/mesa/drivers/dri/radeon/radeon_screen.c
index 3a0a8d208b2..0a1c0bbc38b 100644
--- a/src/mesa/drivers/dri/radeon/radeon_screen.c
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.c
@@ -372,6 +372,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
       screen->drmSupportsFragShader = (sPriv->drmMinor >= 18);
       screen->drmSupportsPointSprites = (sPriv->drmMinor >= 13);
       screen->drmSupportsCubeMapsR100 = (sPriv->drmMinor >= 15);
+      screen->drmSupportsVertexProgram = (sPriv->drmMinor >= 25);
    }
 
    screen->mmio.handle = dri_priv->registerHandle;
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.h b/src/mesa/drivers/dri/radeon/radeon_screen.h
index 3e0f9454e7f..25e6fcf399a 100644
--- a/src/mesa/drivers/dri/radeon/radeon_screen.h
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.h
@@ -98,6 +98,7 @@ typedef struct {
    GLboolean drmSupportsFragShader;     /* need radeon kernel module >= 1.18 */
    GLboolean drmSupportsPointSprites;   /* need radeon kernel module >= 1.13 */
    GLboolean drmSupportsCubeMapsR100;   /* need radeon kernel module >= 1.15 */
+   GLboolean drmSupportsVertexProgram;  /* need radeon kernel module >= 1.25 */
    GLboolean depthHasSurface;
 
    /* Configuration cache with default values for all contexts */
diff --git a/src/mesa/drivers/dri/radeon/server/radeon_reg.h b/src/mesa/drivers/dri/radeon/server/radeon_reg.h
index a1883f644d1..4dcce638460 100644
--- a/src/mesa/drivers/dri/radeon/server/radeon_reg.h
+++ b/src/mesa/drivers/dri/radeon/server/radeon_reg.h
@@ -1883,6 +1883,8 @@
 #       define RADEON_LIGHT_6_SHIFT            0
 #       define RADEON_LIGHT_7_SHIFT            16
 
+#define RADEON_SE_TCL_STATE_FLUSH           0x2284
+
 #define RADEON_SE_TCL_SHININESS             0x2250
 
 #define RADEON_SE_TCL_TEXTURE_PROC_CTL      0x2268