freedreno/a6xx: fix hang with large render target

[mesa.git] / src / gallium / drivers / i915 / i915_fpc_emit.c

diff --git a/src/gallium/drivers/i915/i915_fpc_emit.c b/src/gallium/drivers/i915/i915_fpc_emit.c
index d28595e0fd34c68a73524f0aa5062c7101cfac4c..a4b5ae675a37d6f8ce07bb1ac8b9e608fa9cf186 100644 (file)
--- a/src/gallium/drivers/i915/i915_fpc_emit.c
+++ b/src/gallium/drivers/i915/i915_fpc_emit.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2003 VMware, Inc.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -18,7 +18,7 @@
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
@@ -30,38 +30,6 @@
 #include "i915_fpc.h"
 #include "util/u_math.h"
 
-
-#define A0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
-#define D0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
-#define T0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
-#define A0_SRC0( reg ) (((reg)&UREG_MASK)>>UREG_A0_SRC0_SHIFT_LEFT)
-#define A1_SRC0( reg ) (((reg)&UREG_MASK)<<UREG_A1_SRC0_SHIFT_RIGHT)
-#define A1_SRC1( reg ) (((reg)&UREG_MASK)>>UREG_A1_SRC1_SHIFT_LEFT)
-#define A2_SRC1( reg ) (((reg)&UREG_MASK)<<UREG_A2_SRC1_SHIFT_RIGHT)
-#define A2_SRC2( reg ) (((reg)&UREG_MASK)>>UREG_A2_SRC2_SHIFT_LEFT)
-
-/* These are special, and don't have swizzle/negate bits.
- */
-#define T0_SAMPLER( reg )     (GET_UREG_NR(reg)<<T0_SAMPLER_NR_SHIFT)
-#define T1_ADDRESS_REG( reg ) ((GET_UREG_NR(reg)<<T1_ADDRESS_REG_NR_SHIFT) | \
-                              (GET_UREG_TYPE(reg)<<T1_ADDRESS_REG_TYPE_SHIFT))
-
-
-/* Macros for translating UREG's into the various register fields used
- * by the I915 programmable unit.
- */
-#define UREG_A0_DEST_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_DEST_TYPE_SHIFT)
-#define UREG_A0_SRC0_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_SRC0_TYPE_SHIFT)
-#define UREG_A1_SRC0_SHIFT_RIGHT (A1_SRC0_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
-#define UREG_A1_SRC1_SHIFT_LEFT  (UREG_TYPE_SHIFT - A1_SRC1_TYPE_SHIFT)
-#define UREG_A2_SRC1_SHIFT_RIGHT (A2_SRC1_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
-#define UREG_A2_SRC2_SHIFT_LEFT  (UREG_TYPE_SHIFT - A2_SRC2_TYPE_SHIFT)
-
-#define UREG_MASK         0xffffff00
-#define UREG_TYPE_NR_MASK ((REG_TYPE_MASK << UREG_TYPE_SHIFT) | \
-                          (REG_NR_MASK << UREG_NR_SHIFT))
-
-
 uint
 i915_get_temp(struct i915_fp_compile *p)
 {
@@ -216,16 +184,35 @@ i915_emit_arith(struct i915_fp_compile * p,
  * \param opcode  the instruction opcode
  */
 uint i915_emit_texld( struct i915_fp_compile *p,
-                       uint dest,
-                       uint destmask,
-                       uint sampler,
-                       uint coord,
-                       uint opcode )
+                      uint dest,
+                      uint destmask,
+                      uint sampler,
+                      uint coord,
+                      uint opcode,
+                      uint num_coord )
 {
    const uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
+
    int temp = -1;
+   uint ignore = 0;
+
+   /* Eliminate the useless texture coordinates. Otherwise we end up generating
+    * a swizzle for no reason below. */
+   switch(num_coord) {
+      case 0:
+         ignore |= (0xf << UREG_CHANNEL_X_SHIFT);
+         /* fall-through */
+      case 1:
+         ignore |= (0xf << UREG_CHANNEL_Y_SHIFT);
+         /* fall-through */
+      case 2:
+         ignore |= (0xf << UREG_CHANNEL_Z_SHIFT);
+         /* fall-through */
+      case 3:
+         ignore |= (0xf << UREG_CHANNEL_W_SHIFT);
+   }
 
-   if (coord != k) {
+   if ( (coord & ~ignore ) != (k & ~ignore) ) {
       /* texcoord is swizzled or negated.  Need to allocate a new temporary
        * register (a utemp / unpreserved temp) won't do.
        */
@@ -248,7 +235,7 @@ uint i915_emit_texld( struct i915_fp_compile *p,
    if (destmask != A0_DEST_CHANNEL_ALL) {
       /* if not writing to XYZW... */
       uint tmp = i915_get_utemp(p);
-      i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, opcode );
+      i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, opcode, num_coord );
       i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
       /* XXX release utemp here? */
    }
@@ -369,7 +356,6 @@ i915_emit_const4f(struct i915_fp_compile * p,
 
    // XXX emit swizzle here for 0, 1, -1 and any combination thereof
    // we can use swizzle + neg for that
-   printf("const %f %f %f %f\n",c0,c1,c2,c3);
    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
       if (ifs->constant_flags[reg] == 0xf &&
           ifs->constants[reg][0] == c0 &&