aco: add GFX6/7 subdword lowering tests
authorDaniel Schürmann <daniel@schuermann.dev>
Tue, 26 May 2020 20:56:18 +0000 (21:56 +0100)
committerMarge Bot <eric+marge@anholt.net>
Thu, 30 Jul 2020 16:13:08 +0000 (16:13 +0000)
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Daniel Schürmann <daniel@schuermann.dev>
Acked-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3521>

src/amd/compiler/tests/test_to_hw_instr.cpp

index 075295886a3c9c593f368c78b0b5b0b53831604e..3047a3c54d24d9e40c66c1694280ae851100800d 100644 (file)
 using namespace aco;
 
 BEGIN_TEST(to_hw_instr.swap_subdword)
+   PhysReg v0_lo{256};
+   PhysReg v0_hi{256};
+   PhysReg v0_b1{256};
+   PhysReg v0_b3{256};
+   PhysReg v1_lo{257};
+   PhysReg v1_hi{257};
+   PhysReg v1_b1{257};
+   PhysReg v1_b3{257};
+   PhysReg v2_lo{258};
+   PhysReg v3_lo{259};
+   v0_hi.reg_b += 2;
+   v1_hi.reg_b += 2;
+   v0_b1.reg_b += 1;
+   v1_b1.reg_b += 1;
+   v0_b3.reg_b += 3;
+   v1_b3.reg_b += 3;
+
+   for (unsigned i = GFX6; i <= GFX7; i++) {
+      if (!setup_cs(NULL, (chip_class)i))
+         continue;
+
+      //~gfx[67]>>  p_unit_test 0
+      //~gfx[67]! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
+      //~gfx[67]! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
+      //~gfx[67]! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(0u));
+      bld.pseudo(aco_opcode::p_parallelcopy,
+                 Definition(v0_lo, v2b), Definition(v1_lo, v2b),
+                 Operand(v1_lo, v2b), Operand(v0_lo, v2b));
+
+      //~gfx[67]! p_unit_test 1
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffff, %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_and_b32 0xffff, %0:v[1][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_cvt_pk_u16_u32 %0:v[1][0:16], %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[0] = v_mov_b32 %0:v[1]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(1u));
+      bld.pseudo(aco_opcode::p_create_vector,
+                 Definition(v0_lo, v1),
+                 Operand(v1_lo, v2b), Operand(v0_lo, v2b));
+
+      //~gfx[67]! p_unit_test 2
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffff, %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_and_b32 0xffff, %0:v[1][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_cvt_pk_u16_u32 %0:v[1][0:16], %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[0] = v_mov_b32 %0:v[1]
+      //~gfx[67]! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[2][0:16]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(2u));
+      bld.pseudo(aco_opcode::p_create_vector,
+                 Definition(v0_lo, v6b), Operand(v1_lo, v2b),
+                 Operand(v0_lo, v2b), Operand(v2_lo, v2b));
+
+      //~gfx[67]! p_unit_test 3
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffff, %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_and_b32 0xffff, %0:v[1][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_cvt_pk_u16_u32 %0:v[1][0:16], %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[0] = v_mov_b32 %0:v[1]
+      //~gfx[67]! v1: %0:v[2] = v_and_b32 0xffff, %0:v[2][0:16]
+      //~gfx[67]! v1: %0:v[3] = v_and_b32 0xffff, %0:v[3][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_cvt_pk_u16_u32 %0:v[2][0:16], %0:v[3][0:16]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(3u));
+      bld.pseudo(aco_opcode::p_create_vector,
+                 Definition(v0_lo, v2),
+                 Operand(v1_lo, v2b), Operand(v0_lo, v2b),
+                 Operand(v2_lo, v2b), Operand(v3_lo, v2b));
+
+      //~gfx[67]! p_unit_test 4
+      //~gfx[67]! v1: %0:v[2] = v_and_b32 0xffff, %0:v[2][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_and_b32 0xffff, %0:v[1][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_cvt_pk_u16_u32 %0:v[1][0:16], %0:v[2][0:16]
+      //~gfx[67]! v1: %0:v[3] = v_and_b32 0xffff, %0:v[3][0:16]
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffff, %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[0] = v_cvt_pk_u16_u32 %0:v[0][0:16], %0:v[3][0:16]
+      //~gfx[67]! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
+      //~gfx[67]! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
+      //~gfx[67]! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(4u));
+      bld.pseudo(aco_opcode::p_create_vector,
+                 Definition(v0_lo, v2),
+                 Operand(v1_lo, v2b), Operand(v2_lo, v2b),
+                 Operand(v0_lo, v2b), Operand(v3_lo, v2b));
+
+      //~gfx[67]! p_unit_test 5
+      //~gfx[67]! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][0:16]
+      //~gfx[67]! v2b: %0:v[0][0:16] = v_lshrrev_b32 16, %0:v[1][16:32]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(5u));
+      bld.pseudo(aco_opcode::p_split_vector,
+                 Definition(v1_lo, v2b), Definition(v0_lo, v2b),
+                 Operand(v0_lo, v1));
+
+      //~gfx[67]! p_unit_test 6
+      //~gfx[67]! v2b: %0:v[2][0:16] = v_mov_b32 %0:v[1][0:16]
+      //~gfx[67]! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][0:16]
+      //~gfx[67]! v2b: %0:v[0][0:16] = v_lshrrev_b32 16, %0:v[1][16:32]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(6u));
+      bld.pseudo(aco_opcode::p_split_vector,
+                 Definition(v1_lo, v2b), Definition(v0_lo, v2b),
+                 Definition(v2_lo, v2b), Operand(v0_lo, v6b));
+
+      //~gfx[67]! p_unit_test 7
+      //~gfx[67]! v2b: %0:v[2][0:16] = v_mov_b32 %0:v[1][0:16]
+      //~gfx[67]! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][0:16]
+      //~gfx[67]! v2b: %0:v[0][0:16] = v_lshrrev_b32 16, %0:v[1][16:32]
+      //~gfx[67]! v2b: %0:v[3][0:16] = v_lshrrev_b32 16, %0:v[2][16:32]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(7u));
+      bld.pseudo(aco_opcode::p_split_vector,
+                 Definition(v1_lo, v2b), Definition(v0_lo, v2b),
+                 Definition(v2_lo, v2b), Definition(v3_lo, v2b),
+                 Operand(v0_lo, v2));
+
+      //~gfx[67]! p_unit_test 8
+      //~gfx[67]! v2b: %0:v[2][0:16] = v_lshrrev_b32 16, %0:v[0][16:32]
+      //~gfx[67]! v2b: %0:v[3][0:16] = v_lshrrev_b32 16, %0:v[1][16:32]
+      //~gfx[67]! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
+      //~gfx[67]! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
+      //~gfx[67]! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(8u));
+      bld.pseudo(aco_opcode::p_split_vector,
+                 Definition(v1_lo, v2b), Definition(v2_lo, v2b),
+                 Definition(v0_lo, v2b), Definition(v3_lo, v2b),
+                 Operand(v0_lo, v2));
+
+      //~gfx[67]! p_unit_test 9
+      //~gfx[67]! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
+      //~gfx[67]! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
+      //~gfx[67]! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(9u));
+      bld.pseudo(aco_opcode::p_parallelcopy,
+                 Definition(v0_lo, v1b), Definition(v1_lo, v1b),
+                 Operand(v1_lo, v1b), Operand(v0_lo, v1b));
+
+      //~gfx[67]! p_unit_test 10
+      //~gfx[67]! v1: %0:v[1] = v_and_b32 0xff, %0:v[1][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_lshlrev_b32 8, %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[1] = v_or_b32 %0:v[1][0:8], %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_lshrrev_b32 8, %0:v[0][0:8]
+      //~gfx[67]! v2b: %0:v[0][0:16] = v_mov_b32 %0:v[1][0:16]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(10u));
+      bld.pseudo(aco_opcode::p_create_vector,
+                 Definition(v0_lo, v2b),
+                 Operand(v1_lo, v1b), Operand(v0_lo, v1b));
+
+      //~gfx[67]! p_unit_test 11
+      //~gfx[67]! v1: %0:v[1] = v_and_b32 0xff, %0:v[1][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_lshlrev_b32 8, %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[1] = v_or_b32 %0:v[1][0:8], %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_lshrrev_b32 8, %0:v[0][0:8]
+      //~gfx[67]! v2b: %0:v[0][0:16] = v_mov_b32 %0:v[1][0:16]
+      //~gfx[67]! v1: %0:v[2] = v_and_b32 0xffff, %0:v[2][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffff, %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[0] = v_cvt_pk_u16_u32 %0:v[0][0:16], %0:v[2][0:8]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(11u));
+      bld.pseudo(aco_opcode::p_create_vector,
+                 Definition(v0_lo, v3b), Operand(v1_lo, v1b),
+                 Operand(v0_lo, v1b), Operand(v2_lo, v1b));
+
+      //~gfx[67]! p_unit_test 12
+      //~gfx[67]! v1: %0:v[1] = v_and_b32 0xff, %0:v[1][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_lshlrev_b32 8, %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[1] = v_or_b32 %0:v[1][0:8], %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_lshrrev_b32 8, %0:v[0][0:8]
+      //~gfx[67]! v2b: %0:v[0][0:16] = v_mov_b32 %0:v[1][0:16]
+      //~gfx[67]! v1: %0:v[2] = v_and_b32 0xffff, %0:v[2][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffff, %0:v[0][0:16]
+      //~gfx[67]! v1: %0:v[0] = v_cvt_pk_u16_u32 %0:v[0][0:16], %0:v[2][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffffff, %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[3] = v_lshlrev_b32 24, %0:v[3][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_or_b32 %0:v[0][0:8], %0:v[3][0:8]
+      //~gfx[67]! v1: %0:v[3] = v_lshrrev_b32 24, %0:v[3][0:8]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(12u));
+      bld.pseudo(aco_opcode::p_create_vector,
+                 Definition(v0_lo, v1),
+                 Operand(v1_lo, v1b), Operand(v0_lo, v1b),
+                 Operand(v2_lo, v1b), Operand(v3_lo, v1b));
+
+      //~gfx[67]! p_unit_test 13
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xff, %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_mul_u32_u24 0x101, %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffff, %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_cvt_pk_u16_u32 %0:v[0][0:16], %0:v[0][0:8]
+      //~gfx[67]! v1: %0:v[0] = v_and_b32 0xffffff, %0:v[0][0:8]
+      //~gfx[67]! s1: %0:m0 = s_mov_b32 0x1000001
+      //~gfx[67]! v1: %0:v[0] = v_mul_lo_u32 %0:m0, %0:v[0][0:8]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(13u));
+      Instruction* pseudo = bld.pseudo(aco_opcode::p_create_vector,
+                                       Definition(v0_lo, v1),
+                                       Operand(v0_lo, v1b), Operand(v0_lo, v1b),
+                                       Operand(v0_lo, v1b), Operand(v0_lo, v1b));
+      static_cast<Pseudo_instruction*>(pseudo)->scratch_sgpr = m0;
+
+      //~gfx[67]! p_unit_test 14
+      //~gfx[67]! v1b: %0:v[1][0:8] = v_mov_b32 %0:v[0][0:8]
+      //~gfx[67]! v1b: %0:v[0][0:8] = v_lshrrev_b32 8, %0:v[1][8:16]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(14u));
+      bld.pseudo(aco_opcode::p_split_vector,
+                 Definition(v1_lo, v1b), Definition(v0_lo, v1b),
+                 Operand(v0_lo, v2b));
+
+      //~gfx[67]! p_unit_test 15
+      //~gfx[67]! v1b: %0:v[1][0:8] = v_mov_b32 %0:v[0][0:8]
+      //~gfx[67]! v1b: %0:v[0][0:8] = v_lshrrev_b32 8, %0:v[1][8:16]
+      //~gfx[67]! v1b: %0:v[2][0:8] = v_lshrrev_b32 16, %0:v[1][16:24]
+      //~gfx[67]! v1b: %0:v[3][0:8] = v_lshrrev_b32 24, %0:v[1][24:32]
+      bld.pseudo(aco_opcode::p_unit_test, Operand(15u));
+      bld.pseudo(aco_opcode::p_split_vector,
+                 Definition(v1_lo, v1b), Definition(v0_lo, v1b),
+                 Definition(v2_lo, v1b), Definition(v3_lo, v1b),
+                 Operand(v0_lo, v1));
+
+      //~gfx[67]! s_endpgm
+
+      finish_to_hw_instr_test();
+   }
+
    for (unsigned i = GFX8; i <= GFX9; i++) {
       if (!setup_cs(NULL, (chip_class)i))
          continue;
 
-      PhysReg v0_lo{256};
-      PhysReg v0_hi{256};
-      PhysReg v0_b1{256};
-      PhysReg v0_b3{256};
-      PhysReg v1_lo{257};
-      PhysReg v1_hi{257};
-      PhysReg v1_b1{257};
-      PhysReg v1_b3{257};
-      v0_hi.reg_b += 2;
-      v1_hi.reg_b += 2;
-      v0_b1.reg_b += 1;
-      v1_b1.reg_b += 1;
-      v0_b3.reg_b += 3;
-      v1_b3.reg_b += 3;
-
-      //>> p_unit_test 0
+      //~gfx[89]>> p_unit_test 0
       //~gfx8! v2b: %0:v[0][16:32] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
       //~gfx8! v2b: %0:v[0][0:16] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
       //~gfx8! v2b: %0:v[0][16:32] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
@@ -55,65 +253,65 @@ BEGIN_TEST(to_hw_instr.swap_subdword)
                  Definition(v0_lo, v2b), Definition(v0_hi, v2b),
                  Operand(v0_hi, v2b), Operand(v0_lo, v2b));
 
-      //! p_unit_test 1
+      //~gfx[89]! p_unit_test 1
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
-      //! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][16:32] dst_preserve
+      //~gfx[89]! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][16:32] dst_preserve
       bld.pseudo(aco_opcode::p_unit_test, Operand(1u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_lo, v1), Definition(v1_lo, v2b),
                  Operand(v1_lo, v1), Operand(v0_lo, v2b));
 
-      //! p_unit_test 2
-      //! v2b: %0:v[0][16:32] = v_mov_b32 %0:v[1][16:32] dst_preserve
-      //! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][0:16] dst_preserve
-      //! v2b: %0:v[1][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
-      //! v2b: %0:v[0][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
-      //! v2b: %0:v[1][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
+      //~gfx[89]! p_unit_test 2
+      //~gfx[89]! v2b: %0:v[0][16:32] = v_mov_b32 %0:v[1][16:32] dst_preserve
+      //~gfx[89]! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][0:16] dst_preserve
+      //~gfx[89]! v2b: %0:v[1][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
+      //~gfx[89]! v2b: %0:v[0][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
+      //~gfx[89]! v2b: %0:v[1][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
       bld.pseudo(aco_opcode::p_unit_test, Operand(2u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_lo, v1), Definition(v1_lo, v2b), Definition(v1_hi, v2b),
                  Operand(v1_lo, v1), Operand(v0_lo, v2b), Operand(v0_lo, v2b));
 
-      //! p_unit_test 3
+      //~gfx[89]! p_unit_test 3
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
-      //! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][0:16] dst_preserve
-      //! v1b: %0:v[1][16:24] = v_mov_b32 %0:v[0][16:24] dst_preserve
+      //~gfx[89]! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][0:16] dst_preserve
+      //~gfx[89]! v1b: %0:v[1][16:24] = v_mov_b32 %0:v[0][16:24] dst_preserve
       bld.pseudo(aco_opcode::p_unit_test, Operand(3u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_lo, v1), Definition(v1_b3, v1b),
                  Operand(v1_lo, v1), Operand(v0_b3, v1b));
 
-      //! p_unit_test 4
+      //~gfx[89]! p_unit_test 4
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
-      //! v1b: %0:v[1][8:16] = v_mov_b32 %0:v[0][8:16] dst_preserve
-      //! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][16:32] dst_preserve
+      //~gfx[89]! v1b: %0:v[1][8:16] = v_mov_b32 %0:v[0][8:16] dst_preserve
+      //~gfx[89]! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][16:32] dst_preserve
       bld.pseudo(aco_opcode::p_unit_test, Operand(4u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_lo, v1), Definition(v1_lo, v1b),
                  Operand(v1_lo, v1), Operand(v0_lo, v1b));
 
-      //! p_unit_test 5
+      //~gfx[89]! p_unit_test 5
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[0], %0:v[1]
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[0], %0:v[1]
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[0], %0:v[1]
       //~gfx9! v1: %0:v[1],  v1: %0:v[0] = v_swap_b32 %0:v[0], %0:v[1]
-      //! v1b: %0:v[0][8:16] = v_mov_b32 %0:v[1][8:16] dst_preserve
-      //! v1b: %0:v[0][24:32] = v_mov_b32 %0:v[1][24:32] dst_preserve
+      //~gfx[89]! v1b: %0:v[0][8:16] = v_mov_b32 %0:v[1][8:16] dst_preserve
+      //~gfx[89]! v1b: %0:v[0][24:32] = v_mov_b32 %0:v[1][24:32] dst_preserve
       bld.pseudo(aco_opcode::p_unit_test, Operand(5u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_lo, v1b), Definition(v0_hi, v1b), Definition(v1_lo, v1),
                  Operand(v1_lo, v1b), Operand(v1_hi, v1b), Operand(v0_lo, v1));
 
-      //! p_unit_test 6
+      //~gfx[89]! p_unit_test 6
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
@@ -123,7 +321,7 @@ BEGIN_TEST(to_hw_instr.swap_subdword)
                  Definition(v0_lo, v2b), Definition(v0_hi, v2b), Definition(v1_lo, v1),
                  Operand(v1_lo, v2b), Operand(v1_hi, v2b), Operand(v0_lo, v1));
 
-      //! p_unit_test 7
+      //~gfx[89]! p_unit_test 7
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[0], %0:v[1]
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[0], %0:v[1]
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[0], %0:v[1]
@@ -137,51 +335,51 @@ BEGIN_TEST(to_hw_instr.swap_subdword)
                  Definition(v0_lo, v2b), Definition(v0_hi, v2b), Definition(v1_lo, v1),
                  Operand(v1_hi, v2b), Operand(v1_lo, v2b), Operand(v0_lo, v1));
 
-      //! p_unit_test 8
+      //~gfx[89]! p_unit_test 8
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
-      //! v1b: %0:v[1][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
-      //! v1b: %0:v[0][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
-      //! v1b: %0:v[1][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
+      //~gfx[89]! v1b: %0:v[1][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
+      //~gfx[89]! v1b: %0:v[0][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
+      //~gfx[89]! v1b: %0:v[1][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
       bld.pseudo(aco_opcode::p_unit_test, Operand(8u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_lo, v3b), Definition(v1_lo, v3b),
                  Operand(v1_lo, v3b), Operand(v0_lo, v3b));
 
-      //! p_unit_test 9
+      //~gfx[89]! p_unit_test 9
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
-      //! v1b: %0:v[1][24:32] = v_mov_b32 %0:v[0][24:32] dst_preserve
+      //~gfx[89]! v1b: %0:v[1][24:32] = v_mov_b32 %0:v[0][24:32] dst_preserve
       bld.pseudo(aco_opcode::p_unit_test, Operand(9u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_lo, v3b), Definition(v1_lo, v3b), Definition(v0_b3, v1b),
                  Operand(v1_lo, v3b), Operand(v0_lo, v3b), Operand(v1_b3, v1b));
 
-      //! p_unit_test 10
-      //! v1b: %0:v[1][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
-      //! v1b: %0:v[0][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
-      //! v1b: %0:v[1][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
-      //! v1b: %0:v[1][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
-      //! v1b: %0:v[0][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
-      //! v1b: %0:v[1][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
+      //~gfx[89]! p_unit_test 10
+      //~gfx[89]! v1b: %0:v[1][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
+      //~gfx[89]! v1b: %0:v[0][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
+      //~gfx[89]! v1b: %0:v[1][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
+      //~gfx[89]! v1b: %0:v[1][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
+      //~gfx[89]! v1b: %0:v[0][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
+      //~gfx[89]! v1b: %0:v[1][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
       bld.pseudo(aco_opcode::p_unit_test, Operand(10u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_b1, v2b), Definition(v1_b1, v2b),
                  Operand(v1_b1, v2b), Operand(v0_b1, v2b));
 
-      //! p_unit_test 11
-      //! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][16:32] dst_preserve
-      //! v1: %0:v[0] = v_mov_b32 42
+      //~gfx[89]! p_unit_test 11
+      //~gfx[89]! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][16:32] dst_preserve
+      //~gfx[89]! v1: %0:v[0] = v_mov_b32 42
       bld.pseudo(aco_opcode::p_unit_test, Operand(11u));
       bld.pseudo(aco_opcode::p_parallelcopy,
                  Definition(v0_lo, v1), Definition(v1_lo, v2b),
                  Operand(42u), Operand(v0_hi, v2b));
 
-      //! s_endpgm
+      //~gfx[89]! s_endpgm
 
       finish_to_hw_instr_test();
    }