R5xx: various updates

- fixup VAP_CNTL setup
- remove extra instruction in r5xx passthrough shader
- add notes about pipe config

3 files changed, 56 insertions, 33 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_reg.h b/src/mesa/drivers/dri/r300/r300_reg.h
index 2822b1d4c3a..fee21dae674 100644
--- a/src/mesa/drivers/dri/r300/r300_reg.h
+++ b/src/mesa/drivers/dri/r300/r300_reg.h
@@ -70,6 +70,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
  * Stolen from r200 code from Christoph Brill (It's a guess!)
  */
 #define R300_VAP_CNTL	0x2080
+#       define R300_PVS_NUM_SLOTS_SHIFT                 0
+#       define R300_PVS_NUM_CNTLRS_SHIFT                4
+#       define R300_PVS_NUM_FPUS_SHIFT                  8
+#       define R300_VF_MAX_VTX_NUM_SHIFT                18
+#       define R300_GL_CLIP_SPACE_DEF                   (0 << 22)
+#       define R300_DX_CLIP_SPACE_DEF                   (1 << 22)
+#       define R500_TCL_STATE_OPTIMIZATION              (1 << 23)
 
 /* This register is written directly and also starts data section
  * in many 3d CP_PACKET3's
@@ -375,7 +382,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R300_VAP_CLIP_CNTL                       0x221C
 #       define R300_221C_NORMAL                  0x00000000
 #       define R300_221C_CLEAR                   0x0001C000
-#define R300_VAP_UCP_ENABLE_0 (1 << 0)
+#       define R300_VAP_UCP_ENABLE_0 (1 << 0)
 
 /* These seem to be per-pixel and per-vertex X and Y clipping planes. The first
  * plane is per-pixel and the second plane is per-vertex.
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index 7419b15a55c..0740d7ea4a4 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1981,10 +1981,34 @@ static void r300ResetHwState(r300ContextPtr r300)
 	r300AlphaFunc(ctx, ctx->Color.AlphaFunc, ctx->Color.AlphaRef);
 	r300Enable(ctx, GL_ALPHA_TEST, ctx->Color.AlphaEnabled);
 
-	if (!has_tcl)
-		r300->hw.vap_cntl.cmd[1] = 0x0014045a;
+	/* setup the VAP */
+	/* PVS_NUM_SLOTS, PVS_NUM_CNTLRS, VF_MAX_VTX_NUM need to be adjusted
+	 * dynamically.  PVS_NUM_FPUS is fixed based on asic
+	 */
+	if (has_tcl) {
+	    r300->hw.vap_cntl.cmd[1] = ((10 << R300_PVS_NUM_SLOTS_SHIFT) |
+					(5 << R300_PVS_NUM_CNTLRS_SHIFT) |
+					(12 << R300_VF_MAX_VTX_NUM_SHIFT));
+	    if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
+		r300->hw.vap_cntl.cmd[1] |= R500_TCL_STATE_OPTIMIZATION;
+	} else
+	    r300->hw.vap_cntl.cmd[1] = ((10 << R300_PVS_NUM_SLOTS_SHIFT) |
+					(5 << R300_PVS_NUM_CNTLRS_SHIFT) |
+					(5 << R300_VF_MAX_VTX_NUM_SHIFT));
+
+	if (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV515)
+	    r300->hw.vap_cntl.cmd[1] |= (2 << R300_PVS_NUM_FPUS_SHIFT);
+	else if ((r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV530) ||
+		 (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV560))
+	    r300->hw.vap_cntl.cmd[1] |= (5 << R300_PVS_NUM_FPUS_SHIFT);
+	else if (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R420)
+	    r300->hw.vap_cntl.cmd[1] |= (6 << R300_PVS_NUM_FPUS_SHIFT);
+	else if ((r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R520) ||
+		 (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R580) ||
+		 (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV570))
+	    r300->hw.vap_cntl.cmd[1] |= (8 << R300_PVS_NUM_FPUS_SHIFT);
 	else
-		r300->hw.vap_cntl.cmd[1] = 0x0030045A;	//0x0030065a /* Dangerous */
+	    r300->hw.vap_cntl.cmd[1] |= (4 << R300_PVS_NUM_FPUS_SHIFT);
 
 	r300->hw.vte.cmd[1] = R300_VPORT_X_SCALE_ENA
 	    | R300_VPORT_X_OFFSET_ENA
@@ -2035,20 +2059,27 @@ static void r300ResetHwState(r300ContextPtr r300)
 	r300->hw.gb_misc.cmd[R300_GB_MISC_MSPOS_0] = 0x66666666;
 	r300->hw.gb_misc.cmd[R300_GB_MISC_MSPOS_1] = 0x06666666;
 
-	/* XXX: Other families? */
+	/* num pipes needs to be read back from the GB_PIPE_SELECT register
+	 * on r4xx/r5xx/rs4xx/rs6xx
+	 * should move this to the drm
+	 */
 	r300->hw.gb_misc.cmd[R300_GB_MISC_TILE_CONFIG] =
-	    R300_GB_TILE_ENABLE | R300_GB_TILE_SIZE_16;
+	    R300_GB_TILE_ENABLE | R300_GB_TILE_SIZE_16 /*| R300_GB_SUBPIXEL_1_16*/;
 	switch (r300->radeon.radeonScreen->chip_family) {
 	case CHIP_FAMILY_R300:
 	case CHIP_FAMILY_R350:
-	case CHIP_FAMILY_RV410:
 		r300->hw.gb_misc.cmd[R300_GB_MISC_TILE_CONFIG] |=
 		    R300_GB_TILE_PIPE_COUNT_R300;
 		break;
-	case CHIP_FAMILY_R420:
+	case CHIP_FAMILY_RV350:
 	case CHIP_FAMILY_RV515:
-	case CHIP_FAMILY_R520:
 	case CHIP_FAMILY_RV530:
+	case CHIP_FAMILY_RV410:
+		r300->hw.gb_misc.cmd[R300_GB_MISC_TILE_CONFIG] |=
+		    R300_GB_TILE_PIPE_COUNT_RV300;
+		break;
+	case CHIP_FAMILY_R420:
+	case CHIP_FAMILY_R520:
 	case CHIP_FAMILY_R580:
 	case CHIP_FAMILY_RV560:
 	case CHIP_FAMILY_RV570:
@@ -2329,28 +2360,6 @@ static void r500SetupPixelShader(r300ContextPtr rmesa)
 	/* emit the standard zero shader */
 	R300_STATECHANGE(rmesa, r500fp);
 	i = 1;
-	rmesa->hw.r500fp.cmd[i++] = 0x7808;
-	rmesa->hw.r500fp.cmd[i++] = R500_TEX_ID(0) | R500_TEX_INST_LD | R500_TEX_SEM_ACQUIRE | R500_TEX_IGNORE_UNCOVERED;
-	rmesa->hw.r500fp.cmd[i++] = R500_TEX_SRC_ADDR(0) |  R500_TEX_SRC_S_SWIZ_R |
-		R500_TEX_SRC_T_SWIZ_G |
-		R500_TEX_DST_ADDR(0) |
-		R500_TEX_DST_R_SWIZ_R |
-		R500_TEX_DST_G_SWIZ_G |
-		R500_TEX_DST_B_SWIZ_B |
-		R500_TEX_DST_A_SWIZ_A;
-	rmesa->hw.r500fp.cmd[i++] = R500_DX_ADDR(0) |
-		R500_DX_S_SWIZ_R |
-		R500_DX_T_SWIZ_R |
-		R500_DX_R_SWIZ_R |
-		R500_DX_Q_SWIZ_R |
-		R500_DY_ADDR(0) |
-		R500_DY_S_SWIZ_R |
-		R500_DY_T_SWIZ_R |
-		R500_DY_R_SWIZ_R |
-		R500_DY_Q_SWIZ_R;
-	rmesa->hw.r500fp.cmd[i++] = 0x0;
-	rmesa->hw.r500fp.cmd[i++] = 0x0;
-
 	rmesa->hw.r500fp.cmd[i++] = R500_INST_TYPE_OUT |
 		R500_INST_TEX_SEM_WAIT |
 		R500_INST_LAST |
@@ -2388,7 +2397,7 @@ static void r500SetupPixelShader(r300ContextPtr rmesa)
 		R500_ALU_RGBA_B_SWIZ_0 |
 		R500_ALU_RGBA_A_SWIZ_0;
 
-	bump_r500fp_count(rmesa->hw.r500fp.cmd, 12);
+	bump_r500fp_count(rmesa->hw.r500fp.cmd, 6);
 
 	R300_STATECHANGE(rmesa, r500fp_const);
 	for (i = 0; i < fp->const_nr; i++) {
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.c b/src/mesa/drivers/dri/radeon/radeon_screen.c
index 6f9d9124425..2f57d289fe3 100644
--- a/src/mesa/drivers/dri/radeon/radeon_screen.c
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.c
@@ -649,7 +649,9 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
       screen->chip_flags = RADEON_CHIPSET_TCL;
       break;
 
-   /* RV410 SE chips have half the pipes of regular RV410 */
+      /* RV410 SE chips have half the pipes of regular RV410
+       * Need to get num pipes form the GB_PIPE_SELECT register
+       */
    case PCI_CHIP_RV410_5E4C:
    case PCI_CHIP_RV410_5E4F:
       screen->chip_family = CHIP_FAMILY_RV380;
@@ -709,6 +711,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
    case PCI_CHIP_R520_710E:
    case PCI_CHIP_R520_710F:
       screen->chip_family = CHIP_FAMILY_R520;
+      //screen->chip_flags = RADEON_CHIPSET_TCL;
       fprintf(stderr, "Warning, R520 detected, 3D HAHAHAHAHA!!.\n");
       break;
 
@@ -751,6 +754,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
    case PCI_CHIP_RV515_7210:
    case PCI_CHIP_RV515_7211:
       screen->chip_family = CHIP_FAMILY_RV515;
+      //screen->chip_flags = RADEON_CHIPSET_TCL;
       fprintf(stderr, "Warning, RV515 detected, 3D HAHAHAHAHA!!.\n");
       break;
 
@@ -771,6 +775,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
    case PCI_CHIP_RV530_71DA:
    case PCI_CHIP_RV530_71DE:
       screen->chip_family = CHIP_FAMILY_RV530;
+      //screen->chip_flags = RADEON_CHIPSET_TCL;
       fprintf(stderr, "Warning, RV530 detected, 3D HAHAHAHAHA!!.\n");
       break;
 
@@ -790,6 +795,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
    case PCI_CHIP_R580_724F:
    case PCI_CHIP_R580_7284:
       screen->chip_family = CHIP_FAMILY_R580;
+      //screen->chip_flags = RADEON_CHIPSET_TCL;
       fprintf(stderr, "Warning, R580 detected, 3D HAHAHAHAHA!!.\n");
       break;
 
@@ -806,6 +812,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
    case PCI_CHIP_RV560_7293:
    case PCI_CHIP_RV560_7297:
       screen->chip_family = CHIP_FAMILY_RV560;
+      //screen->chip_flags = RADEON_CHIPSET_TCL;
       fprintf(stderr, "Warning, RV560 detected, 3D HAHAHAHAHA!!.\n");
       break;