amd/registers: fix fields conflict detection

The existing code handled the case where the new definition of the
same field was larger than the old one.
This commit adds a check to handle the reverse case: the new def
is smaller than the old one (= so writing using the merged macro
would affect the next fields).

The affected fields are:
* LGKM_CNT (in SQ_WAVE_IB_STS)
* DONUT_SPLIT (in VGT_TESS_DISTRIBUTION)
* HEAD_QUEUE (in GDS_GWS_RESOURCE)

DONUT_SPLIT is the only one used by radeonsi/radv.

Fixes: e6184b0892 ("amd/registers: scripts for processing register descriptions in JSON")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12063>
(cherry picked from commit 3914bd457b)

.pick_status.json

@@ -1219,7 +1219,7 @@
         "description": "amd/registers: fix fields conflict detection",
         "nominated": true,
         "nomination_type": 1,
-        "resolution": 0,
+        "resolution": 1,
         "main_sha": null,
         "because_sha": "e6184b089240b76942650e847b8a4879821caaa6"
     },

src/amd/registers/makeregheader.py

@@ -112,6 +112,23 @@ def get_chips_comment(chips, parent=None):
 
     return ', '.join(comment)
 
+def detect_conflict(regdb, field_in_type1, field_in_type2):
+    """
+    Returns False if field_in_type1 and field_in_type2 can be merged
+    into a single field = if writing to field_in_type1 bits won't
+    overwrite adjacent fields in type2, and the other way around.
+    """
+    for idx, type_refs in enumerate([field_in_type1.type_refs, field_in_type2.type_refs]):
+        ref = field_in_type2 if idx == 0 else field_in_type1
+        for type_ref in type_refs:
+            for field in regdb.register_type(type_ref).fields:
+                # If a different field in the other type starts in
+                # the tested field's bits[0, 1] interval
+                if (field.bits[0] > ref.bits[0] and
+                    field.bits[0] <= ref.bits[1]):
+                    return True
+
+    return False
 
 class HeaderWriter(object):
     def __init__(self, regdb, guard=None):
@@ -200,21 +217,10 @@ class HeaderWriter(object):
                 if prev.bits[0] != line.bits[0]:
                     continue
 
-                if prev.bits[1] < line.bits[1]:
+                if prev.bits[1] != line.bits[1]:
                     # Current line's field extends beyond the range of prev.
                     # Need to check for conflicts
-                    conflict = False
+                    if detect_conflict(regdb, prev, line):
-                    for type_ref in prev.type_refs:
-                        for field in regdb.register_type(type_ref).fields:
-                            # The only possible conflict is for a prev field
-                            # that starts at a higher bit.
-                            if (field.bits[0] > line.bits[0] and
-                                field.bits[0] <= line.bits[1]):
-                                conflict = True
-                                break
-                        if conflict:
-                            break
-                    if conflict:
                         continue
 
                 prev.bits[1] = max(prev.bits[1], line.bits[1])

src/amd/vulkan/si_cmd_buffer.c

@@ -495,13 +495,13 @@ si_emit_graphics(struct radv_device *device, struct radeon_cmdbuf *cs)
    if (physical_device->rad_info.chip_class >= GFX9) {
       radeon_set_context_reg(cs, R_028B50_VGT_TESS_DISTRIBUTION,
                              S_028B50_ACCUM_ISOLINE(40) | S_028B50_ACCUM_TRI(30) |
-                                S_028B50_ACCUM_QUAD(24) | S_028B50_DONUT_SPLIT(24) |
+                                S_028B50_ACCUM_QUAD(24) | S_028B50_DONUT_SPLIT_GFX9(24) |
                                 S_028B50_TRAP_SPLIT(6));
    } else if (physical_device->rad_info.chip_class >= GFX8) {
       uint32_t vgt_tess_distribution;
 
       vgt_tess_distribution = S_028B50_ACCUM_ISOLINE(32) | S_028B50_ACCUM_TRI(11) |
-                              S_028B50_ACCUM_QUAD(11) | S_028B50_DONUT_SPLIT(16);
+                              S_028B50_ACCUM_QUAD(11) | S_028B50_DONUT_SPLIT_GFX81(16);
 
       if (physical_device->rad_info.family == CHIP_FIJI ||
           physical_device->rad_info.family >= CHIP_POLARIS10)

src/gallium/drivers/radeonsi/si_state.c

@@ -5335,7 +5335,7 @@ void si_init_cs_preamble_state(struct si_context *sctx, bool uses_reg_shadowing)
       unsigned vgt_tess_distribution;
 
       vgt_tess_distribution = S_028B50_ACCUM_ISOLINE(32) | S_028B50_ACCUM_TRI(11) |
-                              S_028B50_ACCUM_QUAD(11) | S_028B50_DONUT_SPLIT(16);
+                              S_028B50_ACCUM_QUAD(11) | S_028B50_DONUT_SPLIT_GFX81(16);
 
       /* Testing with Unigine Heaven extreme tesselation yielded best results
        * with TRAP_SPLIT = 3.
@@ -5362,7 +5362,7 @@ void si_init_cs_preamble_state(struct si_context *sctx, bool uses_reg_shadowing)
 
       si_pm4_set_reg(pm4, R_028B50_VGT_TESS_DISTRIBUTION,
                      S_028B50_ACCUM_ISOLINE(40) | S_028B50_ACCUM_TRI(30) | S_028B50_ACCUM_QUAD(24) |
-                     S_028B50_DONUT_SPLIT(24) | S_028B50_TRAP_SPLIT(6));
+                     S_028B50_DONUT_SPLIT_GFX9(24) | S_028B50_TRAP_SPLIT(6));
       si_pm4_set_reg(pm4, R_028C48_PA_SC_BINNER_CNTL_1,
                      S_028C48_MAX_ALLOC_COUNT(sscreen->info.pbb_max_alloc_count - 1) |
                      S_028C48_MAX_PRIM_PER_BATCH(1023));