aco/tests: add tests for long jumps

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6212>

diff --git a/src/amd/compiler/tests/test_assembler.cpp b/src/amd/compiler/tests/test_assembler.cpp
index eb3756a0cf503b6965f5373121439c2d8c203aef..ae4247e49abb85107d8ed71afd8c4b6b449d7ac0 100644 (file)
--- a/src/amd/compiler/tests/test_assembler.cpp
+++ b/src/amd/compiler/tests/test_assembler.cpp
@@ -58,3 +58,172 @@ BEGIN_TEST(assembler.branch_3f)
 
    finish_assembler_test();
 END_TEST
+
+BEGIN_TEST(assembler.long_jump.unconditional_forwards)
+   if (!setup_cs(NULL, (chip_class)GFX10))
+      return;
+
+   //!BB0:
+   //! s_getpc_b64 s[0:1]                                          ; be801f00
+   //! s_addc_u32 s0, s0, 0x20018                                  ; 8200ff00 00020018
+   //! s_addc_u32 s1, s1, 0                                        ; 82018001
+   //! s_bitcmp1_b32 s0, 0                                         ; bf0d8000
+   //! s_bitset0_b32 s0, 0                                         ; be801b80
+   //! s_setpc_b64 s[0:1]                                          ; be802000
+   bld.sopp(aco_opcode::s_branch, Definition(PhysReg(0), s2), 2);
+
+   bld.reset(program->create_and_insert_block());
+
+   //! s_nop 0                                                     ; bf800000
+   //!(then repeated 32767 times)
+   for (unsigned i = 0; i < INT16_MAX + 1; i++)
+      bld.sopp(aco_opcode::s_nop, -1, 0);
+
+   //! BB2:
+   //! s_endpgm                                                    ; bf810000
+   bld.reset(program->create_and_insert_block());
+
+   program->blocks[2].linear_preds.push_back(0u);
+   program->blocks[2].linear_preds.push_back(1u);
+
+   finish_assembler_test();
+END_TEST
+
+BEGIN_TEST(assembler.long_jump.conditional_forwards)
+   if (!setup_cs(NULL, (chip_class)GFX10))
+      return;
+
+   //! BB0:
+   //! s_cbranch_scc1 BB1                                          ; bf850007
+   //! s_getpc_b64 s[0:1]                                          ; be801f00
+   //! s_addc_u32 s0, s0, 0x20018                                  ; 8200ff00 00020018
+   //! s_addc_u32 s1, s1, 0                                        ; 82018001
+   //! s_bitcmp1_b32 s0, 0                                         ; bf0d8000
+   //! s_bitset0_b32 s0, 0                                         ; be801b80
+   //! s_setpc_b64 s[0:1]                                          ; be802000
+   bld.sopp(aco_opcode::s_cbranch_scc0, Definition(PhysReg(0), s2), 2);
+
+   bld.reset(program->create_and_insert_block());
+
+   //! BB1:
+   //! s_nop 0 ; bf800000
+   //!(then repeated 32767 times)
+   for (unsigned i = 0; i < INT16_MAX + 1; i++)
+      bld.sopp(aco_opcode::s_nop, -1, 0);
+
+   //! BB2:
+   //! s_endpgm                                                    ; bf810000
+   bld.reset(program->create_and_insert_block());
+
+   program->blocks[1].linear_preds.push_back(0u);
+   program->blocks[2].linear_preds.push_back(0u);
+   program->blocks[2].linear_preds.push_back(1u);
+
+   finish_assembler_test();
+END_TEST
+
+BEGIN_TEST(assembler.long_jump.unconditional_backwards)
+   if (!setup_cs(NULL, (chip_class)GFX10))
+      return;
+
+   //!BB0:
+   //! s_nop 0                                                     ; bf800000
+   //!(then repeated 32767 times)
+   for (unsigned i = 0; i < INT16_MAX + 1; i++)
+      bld.sopp(aco_opcode::s_nop, -1, 0);
+
+   //! s_getpc_b64 s[0:1]                                          ; be801f00
+   //! s_addc_u32 s0, s0, 0xfffdfffc                               ; 8200ff00 fffdfffc
+   //! s_addc_u32 s1, s1, -1                                       ; 8201c101
+   //! s_bitcmp1_b32 s0, 0                                         ; bf0d8000
+   //! s_bitset0_b32 s0, 0                                         ; be801b80
+   //! s_setpc_b64 s[0:1]                                          ; be802000
+   bld.sopp(aco_opcode::s_branch, Definition(PhysReg(0), s2), 0);
+
+   //! BB1:
+   //! s_endpgm                                                    ; bf810000
+   bld.reset(program->create_and_insert_block());
+
+   program->blocks[0].linear_preds.push_back(0u);
+   program->blocks[1].linear_preds.push_back(0u);
+
+   finish_assembler_test();
+END_TEST
+
+BEGIN_TEST(assembler.long_jump.conditional_backwards)
+   if (!setup_cs(NULL, (chip_class)GFX10))
+      return;
+
+   //!BB0:
+   //! s_nop 0                                                     ; bf800000
+   //!(then repeated 32767 times)
+   for (unsigned i = 0; i < INT16_MAX + 1; i++)
+      bld.sopp(aco_opcode::s_nop, -1, 0);
+
+   //! s_cbranch_execz BB1                                         ; bf880007
+   //! s_getpc_b64 s[0:1]                                          ; be801f00
+   //! s_addc_u32 s0, s0, 0xfffdfff8                               ; 8200ff00 fffdfff8
+   //! s_addc_u32 s1, s1, -1                                       ; 8201c101
+   //! s_bitcmp1_b32 s0, 0                                         ; bf0d8000
+   //! s_bitset0_b32 s0, 0                                         ; be801b80
+   //! s_setpc_b64 s[0:1]                                          ; be802000
+   bld.sopp(aco_opcode::s_cbranch_execnz, Definition(PhysReg(0), s2), 0);
+
+   //! BB1:
+   //! s_endpgm                                                    ; bf810000
+   bld.reset(program->create_and_insert_block());
+
+   program->blocks[0].linear_preds.push_back(0u);
+   program->blocks[1].linear_preds.push_back(0u);
+
+   finish_assembler_test();
+END_TEST
+
+BEGIN_TEST(assembler.long_jump.3f)
+   if (!setup_cs(NULL, (chip_class)GFX10))
+      return;
+
+   //! BB0:
+   //! s_branch BB1                                                ; bf820040
+   //! s_nop 0                                                     ; bf800000
+   bld.sopp(aco_opcode::s_branch, Definition(PhysReg(0), s2), 1);
+
+   for (unsigned i = 0; i < 0x3f - 7; i++) // a unconditional long jump is 7 dwords
+      bld.vop1(aco_opcode::v_nop);
+   bld.sopp(aco_opcode::s_branch, Definition(PhysReg(0), s2), 2);
+
+   bld.reset(program->create_and_insert_block());
+   for (unsigned i = 0; i < INT16_MAX + 1; i++)
+      bld.vop1(aco_opcode::v_nop);
+   bld.reset(program->create_and_insert_block());
+
+   program->blocks[1].linear_preds.push_back(0u);
+   program->blocks[2].linear_preds.push_back(0u);
+   program->blocks[2].linear_preds.push_back(1u);
+
+   finish_assembler_test();
+END_TEST
+
+BEGIN_TEST(assembler.long_jump.constaddr)
+   if (!setup_cs(NULL, (chip_class)GFX10))
+      return;
+
+   //>> s_getpc_b64 s[0:1]                                          ; be801f00
+   bld.sopp(aco_opcode::s_branch, Definition(PhysReg(0), s2), 2);
+
+   bld.reset(program->create_and_insert_block());
+
+   for (unsigned i = 0; i < INT16_MAX + 1; i++)
+      bld.sopp(aco_opcode::s_nop, -1, 0);
+
+   bld.reset(program->create_and_insert_block());
+
+   //>> s_getpc_b64 s[0:1]                                          ; be801f00
+   //! s_add_u32 s0, s0, 0xe0                                      ; 8000ff00 000000e0
+   bld.sop1(aco_opcode::p_constaddr, Definition(PhysReg(0), s2), Operand(0u));
+
+   program->blocks[2].linear_preds.push_back(0u);
+   program->blocks[2].linear_preds.push_back(1u);
+
+   finish_assembler_test();
+END_TEST