aco: don't encode src2 for v_writelane_b32_e64

Encoding src2 doesn't cause issues for print_asm() because we have a
workaround there, but it does for RGP and it seems the developers are not
interested in fixing it.

https://github.com/GPUOpen-Tools/radeon_gpu_profiler/issues/61

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Tested-by: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14832>

diff --git a/src/amd/compiler/aco_assembler.cpp b/src/amd/compiler/aco_assembler.cpp
index 4c404ba..3095f0b 100644 (file)
--- a/src/amd/compiler/aco_assembler.cpp
+++ b/src/amd/compiler/aco_assembler.cpp
@@ -625,6 +625,10 @@ emit_instruction(asm_context& ctx, std::vector<uint32_t>& out, Instruction* inst
          encoding = 0;
          if (instr->opcode == aco_opcode::v_interp_mov_f32) {
             encoding = 0x3 & instr->operands[0].constantValue();
+         } else if (instr->opcode == aco_opcode::v_writelane_b32_e64) {
+            encoding |= instr->operands[0].physReg() << 0;
+            encoding |= instr->operands[1].physReg() << 9;
+            /* Encoding src2 works fine with hardware but breaks some disassemblers. */
          } else {
             for (unsigned i = 0; i < instr->operands.size(); i++)
                encoding |= instr->operands[i].physReg() << (i * 9);

diff --git a/src/amd/compiler/aco_print_asm.cpp b/src/amd/compiler/aco_print_asm.cpp
index a2e88b7..673f415 100644 (file)
--- a/src/amd/compiler/aco_print_asm.cpp
+++ b/src/amd/compiler/aco_print_asm.cpp
@@ -271,12 +271,6 @@ std::pair<bool, size_t>
 disasm_instr(chip_class chip, LLVMDisasmContextRef disasm, uint32_t* binary, unsigned exec_size,
              size_t pos, char* outline, unsigned outline_size)
 {
-   /* mask out src2 on v_writelane_b32 */
-   if (((chip == GFX8 || chip == GFX9) && (binary[pos] & 0xffff8000) == 0xd28a0000) ||
-       (chip >= GFX10 && (binary[pos] & 0xffff8000) == 0xd7610000)) {
-      binary[pos + 1] = binary[pos + 1] & 0xF803FFFF;
-   }
-
    size_t l =
       LLVMDisasmInstruction(disasm, (uint8_t*)&binary[pos], (exec_size - pos) * sizeof(uint32_t),
                             pos * 4, outline, outline_size);