From f267e7797fc5bb12367c8a151d3f976380a49c57 Mon Sep 17 00:00:00 2001
From: Job Noorman <jnoorman@igalia.com>
Date: Tue, 5 May 2026 16:16:04 +0200
Subject: [PATCH] ir3: allow const src0 for ldg.a/stg.a/ray_intersection

While they allow a const src, they don't allow the full range of const
registers since the encoding only uses 8 bits. Add a new helper
(ir3_valid_const) that indicates this and use that in ir3_cp.

Totals from 2278 (1.29% of 176266) affected shaders:
MaxWaves: 30124 -> 30376 (+0.84%)
Instrs: 1472806 -> 1468750 (-0.28%); split: -0.44%, +0.16%
CodeSize: 2990430 -> 2967568 (-0.76%); split: -0.88%, +0.11%
NOPs: 244627 -> 246385 (+0.72%); split: -1.71%, +2.43%
MOVs: 51408 -> 45667 (-11.17%); split: -13.16%, +1.99%
Full: 35639 -> 35216 (-1.19%)
(ss): 46519 -> 42591 (-8.44%); split: -12.16%, +3.71%
(sy): 16763 -> 17269 (+3.02%); split: -1.35%, +4.37%
(ss)-stall: 187784 -> 180294 (-3.99%); split: -9.06%, +5.07%
(sy)-stall: 922477 -> 936696 (+1.54%); split: -3.40%, +4.94%
Cat0: 268783 -> 270558 (+0.66%); split: -1.54%, +2.20%
Cat1: 63982 -> 58277 (-8.92%); split: -10.59%, +1.68%
Cat2: 523798 -> 523707 (-0.02%)
Cat7: 47476 -> 47441 (-0.07%); split: -0.11%, +0.03%

Signed-off-by: Job Noorman <jnoorman@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41365>
---
 src/freedreno/ir3/ir3.c    | 25 +++++++++++++++++++++++++
 src/freedreno/ir3/ir3.h    |  2 ++
 src/freedreno/ir3/ir3_cp.c |  5 +++++
 3 files changed, 32 insertions(+)

diff --git a/src/freedreno/ir3/ir3.c b/src/freedreno/ir3/ir3.c
index bc493a9dc76..7dc042c9a55 100644
--- a/src/freedreno/ir3/ir3.c
+++ b/src/freedreno/ir3/ir3.c
@@ -1837,6 +1837,12 @@ ir3_valid_flags(struct ir3_instruction *instr, unsigned n, unsigned flags)
          else if (n == 1)
             valid_flags |= IR3_REG_SHARED;
       }
+      if (compiler->gen >= 7 &&
+          (instr->opc == OPC_LDG_A || instr->opc == OPC_STG_A ||
+           instr->opc == OPC_RAY_INTERSECTION) &&
+          n == 0) {
+         valid_flags |= IR3_REG_CONST;
+      }
 
       if (flags & ~valid_flags)
          return false;
@@ -1964,6 +1970,25 @@ ir3_valid_immediate(struct ir3_instruction *instr, int32_t immed)
    return !(immed & ~0x1ff) || !(-(uint32_t)immed & ~0x1ff);
 }
 
+/* Some instructions (e.g., cat6) don't support the full range of const
+ * registers as src.
+ */
+bool
+ir3_valid_const(struct ir3_instruction *instr, unsigned src_n, unsigned num)
+{
+   assert(ir3_valid_flags(instr, src_n, IR3_REG_CONST));
+
+   switch (instr->opc) {
+   case OPC_LDG_A:
+   case OPC_STG_A:
+   case OPC_RAY_INTERSECTION:
+      assert(src_n == 0);
+      return num < (1 << 8);
+   default:
+      return true;
+   }
+}
+
 struct ir3_instruction *
 ir3_get_cond_for_nonzero_compare(struct ir3_instruction *instr)
 {
diff --git a/src/freedreno/ir3/ir3.h b/src/freedreno/ir3/ir3.h
index 9d5351e6386..83517828918 100644
--- a/src/freedreno/ir3/ir3.h
+++ b/src/freedreno/ir3/ir3.h
@@ -998,6 +998,8 @@ int ir3_flut(struct ir3_register *src_reg);
 bool ir3_valid_flags(struct ir3_instruction *instr, unsigned n, unsigned flags);
 
 bool ir3_valid_immediate(struct ir3_instruction *instr, int32_t immed);
+bool ir3_valid_const(struct ir3_instruction *instr, unsigned src_n,
+                     unsigned num);
 
 /**
  * Given an instruction whose result we want to test for nonzero, return a
diff --git a/src/freedreno/ir3/ir3_cp.c b/src/freedreno/ir3/ir3_cp.c
index 7647265828d..6fd9955fe1d 100644
--- a/src/freedreno/ir3/ir3_cp.c
+++ b/src/freedreno/ir3/ir3_cp.c
@@ -409,6 +409,11 @@ reg_cp(struct ir3_cp_ctx *ctx, struct ir3_instruction *instr,
        * dependency.
        */
       if (src_reg->flags & IR3_REG_CONST) {
+         if (!(src_reg->flags & IR3_REG_RELATIV) &&
+             !ir3_valid_const(instr, n, src_reg->num)) {
+            return false;
+         }
+
          /* an instruction cannot reference two different
           * address registers:
           */