KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

aco: fix p_insert lowering with 16bit sources 

The previous lowering only wrote a single byte.

Fixes: 2f94353735 ('aco: add p_extract/p_insert')
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12640>
This commit is contained in:
Daniel Schürmann 2021-09-01 15:22:12 +02:00
parent cc62fbed6d
commit cc4682ed47
2 changed files with 9 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
src/amd/compiler/aco_lower_to_hw_instr.cpp
View File

@ -2153,16 +2153,11 @@ lower_to_hw_instr(Program* program)
}
} else {
assert(dst.regClass() == v2b);
aco_ptr<SDWA_instruction> sdwa{create_instruction<SDWA_instruction>(
aco_opcode::v_mov_b32,
(Format)((uint16_t)Format::VOP1 | (uint16_t)Format::SDWA), 1, 1)};
sdwa->operands[0] = op;
sdwa->definitions[0] =
Definition(dst.physReg().advance(-dst.physReg().byte()), v1);
sdwa->sel[0] = sdwa_uword;
sdwa->dst_sel = sdwa_ubyte0 + dst.physReg().byte() + index;
sdwa->dst_preserve = 1;
bld.insert(std::move(sdwa));
Operand sdwa_op = Operand(op.physReg().advance(-op.physReg().byte()),
RegClass::get(op.regClass().type(), 4));
bld.vop2_sdwa(aco_opcode::v_lshlrev_b32, dst, Operand::c32(offset), sdwa_op)
.instr->sdwa()
.sel[1] = sdwa_ubyte0 + op.physReg().byte();
}
break;
}

8
src/amd/compiler/tests/test_to_hw_instr.cpp
View File

@ -640,15 +640,15 @@ BEGIN_TEST(to_hw_instr.insert)
//>> p_unit_test 2
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(2u));
//~gfx7! v2b: %_:v[0][0:16] = v_bfe_u32 %_:v[1][0:16], 0, 8
//~gfx[^7]! v1: %_:v[0] = v_mov_b32 %_:v[1][0:16] dst_sel:ubyte0 dst_preserve
//~gfx[^7]! v2b: %0:v[0][0:16] = v_lshlrev_b32 0, %0:v[1][0:7] dst_preserve
INS(0, 0)
//~gfx[^7]! v1: %_:v[0] = v_mov_b32 %_:v[1][0:16] dst_sel:ubyte2 dst_preserve
//~gfx[^7]! v2b: %0:v[0][16:32] = v_lshlrev_b32 0, %0:v[1][0:7] dst_preserve
if (i != GFX7)
INS(0, 2)
//~gfx7! v2b: %_:v[0][0:16] = v_lshlrev_b32 8, %_:v[1][0:16]
//~gfx[^7]! v1: %_:v[0] = v_mov_b32 %_:v[1][0:16] dst_sel:ubyte1 dst_preserve
//~gfx[^7]! v2b: %0:v[0][0:16] = v_lshlrev_b32 8, %0:v[1][0:7] dst_preserve
INS(1, 0)
//~gfx[^7]! v1: %_:v[0] = v_mov_b32 %_:v[1][0:16] dst_sel:ubyte3 dst_preserve
//~gfx[^7]! v2b: %0:v[0][16:32] = v_lshlrev_b32 8, %0:v[1][0:7] dst_preserve
if (i != GFX7)
INS(1, 2)