broadcom/compiler: emit TMU flush before a jump

Like in the case of emitting a block, process pending TMU operations
before a jump is executed.

Fixes dEQP-VK.graphicsfuzz.stable-binarysearch-tree-nested-if-and-conditional.

Fixes: 197090a3fc ("broadcom/compiler: implement pipelining for general
TMU operations")

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11971>

src/broadcom/ci/deqp-v3dv-rpi4-skips.txt

@@ -2,9 +2,6 @@
 dEQP-VK.draw.inverted_depth_ranges.nodepthclamp_deltazero
 dEQP-VK.rasterization.depth_bias.d32_sfloat
 
-# This test causes GPU hangs (vk-gl-cts 1.2.6.2)
-dEQP-VK.graphicsfuzz.stable-binarysearch-tree-nested-if-and-conditional
-
 # Timeout tests (> 1 minute to run)
 dEQP-VK.api.object_management.max_concurrent.query_pool
 dEQP-VK.graphicsfuzz.spv-stable-maze-flatten-copy-composite

src/broadcom/compiler/nir_to_vir.c

@@ -3601,6 +3601,10 @@ ntq_emit_instr(struct v3d_compile *c, nir_instr *instr)
                 break;
 
         case nir_instr_type_jump:
+                /* Always flush TMU before jumping to another block, for the
+                 * same reasons as in ntq_emit_block.
+                 */
+                ntq_flush_tmu(c);
                 if (vir_in_nonuniform_control_flow(c))
                         ntq_emit_jump(c, nir_instr_as_jump(instr));
                 else