aco: fix v_or(s_lshl) and v_add(s_lshl) optimizations

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Fixes: d1621834f367d41500b7c1a819c046eb429fb8a6
    ('aco: combine VALU and SALU into various VOP3 instructions')

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2822
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4717>

diff --git a/src/amd/compiler/aco_optimizer.cpp b/src/amd/compiler/aco_optimizer.cpp
index 0b4729e512d9591c10dd521df319370d150622d2..7c9eb3e1100d34f1e1aceadb0e7d7e113ea8b447 100644 (file)
--- a/src/amd/compiler/aco_optimizer.cpp
+++ b/src/amd/compiler/aco_optimizer.cpp
@@ -2475,7 +2475,7 @@ void combine_instruction(opt_ctx &ctx, Block& block, aco_ptr<Instruction>& instr
       else if (combine_three_valu_op(ctx, instr, aco_opcode::v_or_b32, aco_opcode::v_or3_b32, "012", 1 | 2)) ;
       else if (combine_three_valu_op(ctx, instr, aco_opcode::s_and_b32, aco_opcode::v_and_or_b32, "120", 1 | 2)) ;
       else if (combine_three_valu_op(ctx, instr, aco_opcode::v_and_b32, aco_opcode::v_and_or_b32, "120", 1 | 2)) ;
-      else if (combine_three_valu_op(ctx, instr, aco_opcode::s_lshl_b32, aco_opcode::v_lshl_or_b32, "201", 1 | 2)) ;
+      else if (combine_three_valu_op(ctx, instr, aco_opcode::s_lshl_b32, aco_opcode::v_lshl_or_b32, "120", 1 | 2)) ;
       else combine_three_valu_op(ctx, instr, aco_opcode::v_lshlrev_b32, aco_opcode::v_lshl_or_b32, "210", 1 | 2);
    } else if (instr->opcode == aco_opcode::v_add_u32 && ctx.program->chip_class >= GFX9) {
       if (combine_three_valu_op(ctx, instr, aco_opcode::s_xor_b32, aco_opcode::v_xad_u32, "120", 1 | 2)) ;
@@ -2483,7 +2483,7 @@ void combine_instruction(opt_ctx &ctx, Block& block, aco_ptr<Instruction>& instr
       else if (combine_three_valu_op(ctx, instr, aco_opcode::s_add_i32, aco_opcode::v_add3_u32, "012", 1 | 2)) ;
       else if (combine_three_valu_op(ctx, instr, aco_opcode::s_add_u32, aco_opcode::v_add3_u32, "012", 1 | 2)) ;
       else if (combine_three_valu_op(ctx, instr, aco_opcode::v_add_u32, aco_opcode::v_add3_u32, "012", 1 | 2)) ;
-      else if (combine_three_valu_op(ctx, instr, aco_opcode::s_lshl_b32, aco_opcode::v_lshl_add_u32, "201", 1 | 2)) ;
+      else if (combine_three_valu_op(ctx, instr, aco_opcode::s_lshl_b32, aco_opcode::v_lshl_add_u32, "120", 1 | 2)) ;
       else combine_three_valu_op(ctx, instr, aco_opcode::v_lshlrev_b32, aco_opcode::v_lshl_add_u32, "210", 1 | 2);
    } else if (instr->opcode == aco_opcode::v_lshlrev_b32 && ctx.program->chip_class >= GFX9) {
       combine_three_valu_op(ctx, instr, aco_opcode::v_add_u32, aco_opcode::v_add_lshl_u32, "120", 2);