turnip: Allow nested CP_COND_REG_EXEC

This ends up being needed for moving tile loads into the draw cs.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16826>

4 files changed, 24 insertions, 15 deletions

diff --git a/src/freedreno/vulkan/tu_cmd_buffer.c b/src/freedreno/vulkan/tu_cmd_buffer.c
index 8945bbf7681..93fab52fda8 100644
--- a/src/freedreno/vulkan/tu_cmd_buffer.c
+++ b/src/freedreno/vulkan/tu_cmd_buffer.c
@@ -160,7 +160,7 @@ tu_emit_cache_flush_ccu(struct tu_cmd_buffer *cmd_buffer,
 
    assert(ccu_state != TU_CMD_CCU_UNKNOWN);
    /* It's unsafe to flush inside condition because we clear flush_bits */
-   assert(!cs->cond_flags);
+   assert(!cs->cond_stack_depth);
 
    /* Changing CCU state must involve invalidating the CCU. In sysmem mode,
     * the CCU may also contain data that we haven't flushed out yet, so we
diff --git a/src/freedreno/vulkan/tu_cs.c b/src/freedreno/vulkan/tu_cs.c
index 73b80758ca9..1ff85734c04 100644
--- a/src/freedreno/vulkan/tu_cs.c
+++ b/src/freedreno/vulkan/tu_cs.c
@@ -376,9 +376,9 @@ tu_cs_reserve_space(struct tu_cs *cs, uint32_t reserved_size)
          tu_cs_add_entry(cs);
       }
 
-      if (cs->cond_flags) {
+      for (uint32_t i = 0; i < cs->cond_stack_depth; i++) {
          /* Subtract one here to account for the DWORD field itself. */
-         *cs->cond_dwords = cs->cur - cs->cond_dwords - 1;
+         *cs->cond_dwords[i] = cs->cur - cs->cond_dwords[i] - 1;
 
          /* space for CP_COND_REG_EXEC in next bo */
          reserved_size += 3;
@@ -390,14 +390,16 @@ tu_cs_reserve_space(struct tu_cs *cs, uint32_t reserved_size)
       if (result != VK_SUCCESS)
          return result;
 
-      /* if inside a condition, emit a new CP_COND_REG_EXEC */
-      if (cs->cond_flags) {
+      if (cs->cond_stack_depth) {
          cs->reserved_end = cs->cur + reserved_size;
+      }
 
+      /* Re-emit CP_COND_REG_EXECs */
+      for (uint32_t i = 0; i < cs->cond_stack_depth; i++) {
          tu_cs_emit_pkt7(cs, CP_COND_REG_EXEC, 2);
-         tu_cs_emit(cs, cs->cond_flags);
+         tu_cs_emit(cs, cs->cond_flags[i]);
 
-         cs->cond_dwords = cs->cur;
+         cs->cond_dwords[i] = cs->cur;
 
          /* Emit dummy DWORD field here */
          tu_cs_emit(cs, CP_COND_REG_EXEC_1_DWORDS(0));
diff --git a/src/freedreno/vulkan/tu_cs.h b/src/freedreno/vulkan/tu_cs.h
index 1a01f76a8cf..892c37a3294 100644
--- a/src/freedreno/vulkan/tu_cs.h
+++ b/src/freedreno/vulkan/tu_cs.h
@@ -290,16 +290,18 @@ static inline void
 tu_cond_exec_start(struct tu_cs *cs, uint32_t cond_flags)
 {
    assert(cs->mode == TU_CS_MODE_GROW);
-   assert(!cs->cond_flags && cond_flags);
+   assert(cs->cond_stack_depth < TU_COND_EXEC_STACK_SIZE);
 
    tu_cs_emit_pkt7(cs, CP_COND_REG_EXEC, 2);
    tu_cs_emit(cs, cond_flags);
 
-   cs->cond_flags = cond_flags;
-   cs->cond_dwords = cs->cur;
+   cs->cond_flags[cs->cond_stack_depth] = cond_flags;
+   cs->cond_dwords[cs->cond_stack_depth] = cs->cur;
 
    /* Emit dummy DWORD field here */
    tu_cs_emit(cs, CP_COND_REG_EXEC_1_DWORDS(0));
+
+   cs->cond_stack_depth++;
 }
 #define CP_COND_EXEC_0_RENDER_MODE_GMEM \
    (CP_COND_REG_EXEC_0_MODE(RENDER_MODE) | CP_COND_REG_EXEC_0_GMEM)
@@ -309,11 +311,13 @@ tu_cond_exec_start(struct tu_cs *cs, uint32_t cond_flags)
 static inline void
 tu_cond_exec_end(struct tu_cs *cs)
 {
-   assert(cs->cond_flags);
+   assert(cs->cond_stack_depth > 0);
+   cs->cond_stack_depth--;
 
-   cs->cond_flags = 0;
+   cs->cond_flags[cs->cond_stack_depth] = 0;
    /* Subtract one here to account for the DWORD field itself. */
-   *cs->cond_dwords = cs->cur - cs->cond_dwords - 1;
+   *cs->cond_dwords[cs->cond_stack_depth] =
+      cs->cur - cs->cond_dwords[cs->cond_stack_depth] - 1;
 }
 
 #define fd_reg_pair tu_reg_value
diff --git a/src/freedreno/vulkan/tu_private.h b/src/freedreno/vulkan/tu_private.h
index 0fca4120d8e..2ad2d6e6511 100644
--- a/src/freedreno/vulkan/tu_private.h
+++ b/src/freedreno/vulkan/tu_private.h
@@ -764,6 +764,8 @@ enum tu_cs_mode
    TU_CS_MODE_SUB_STREAM,
 };
 
+#define TU_COND_EXEC_STACK_SIZE 4
+
 struct tu_cs
 {
    uint32_t *start;
@@ -787,8 +789,9 @@ struct tu_cs
    struct tu_bo *refcount_bo;
 
    /* state for cond_exec_start/cond_exec_end */
-   uint32_t cond_flags;
-   uint32_t *cond_dwords;
+   uint32_t cond_stack_depth;
+   uint32_t cond_flags[TU_COND_EXEC_STACK_SIZE];
+   uint32_t *cond_dwords[TU_COND_EXEC_STACK_SIZE];
 };
 
 struct tu_device_memory