continue;
//>> Validation results:
- //~gfx7! SDWA is GFX8+ only: v1: %t0 = v_mul_f32 %a, %b
+ //~gfx7! SDWA is GFX8+ only: v1: %t0 = v_mul_f32 %a, %b dst_sel:dword src0_sel:dword src1_sel:dword
//~gfx7! Validation failed
//~gfx([89]|10)! Validation passed
bld.vop2_sdwa(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
continue;
//>> Validation results:
- //~gfx8! Wrong source position for SGPR argument: v1: %_ = v_mul_f32 %sgpr0, %vgpr1
- //~gfx8! Wrong source position for SGPR argument: v1: %_ = v_mul_f32 %vgpr0, %sgpr1
+ //~gfx8! Wrong source position for SGPR argument: v1: %_ = v_mul_f32 %sgpr0, %vgpr1 dst_sel:dword src0_sel:dword src1_sel:dword
+ //~gfx8! Wrong source position for SGPR argument: v1: %_ = v_mul_f32 %vgpr0, %sgpr1 dst_sel:dword src0_sel:dword src1_sel:dword
bld.vop2_sdwa(aco_opcode::v_mul_f32, bld.def(v1), inputs[2], inputs[1]);
bld.vop2_sdwa(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[3]);
- //~gfx8! Wrong source position for constant argument: v1: %_ = v_mul_f32 4, %vgpr1
- //~gfx8! Wrong source position for constant argument: v1: %_ = v_mul_f32 %vgpr0, 4
+ //~gfx8! Wrong source position for constant argument: v1: %_ = v_mul_f32 4, %vgpr1 dst_sel:dword src0_sel:dword src1_sel:dword
+ //~gfx8! Wrong source position for constant argument: v1: %_ = v_mul_f32 %vgpr0, 4 dst_sel:dword src0_sel:dword src1_sel:dword
bld.vop2_sdwa(aco_opcode::v_mul_f32, bld.def(v1), Operand::c32(4u), inputs[1]);
bld.vop2_sdwa(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], Operand::c32(4u));
- //! Literal applied on wrong instruction format: v1: %_ = v_mul_f32 0x1234, %vgpr1
- //! Literal applied on wrong instruction format: v1: %_ = v_mul_f32 %vgpr0, 0x1234
- //! Wrong source position for Literal argument: v1: %_ = v_mul_f32 %vgpr0, 0x1234
+ //! Literal applied on wrong instruction format: v1: %_ = v_mul_f32 0x1234, %vgpr1 dst_sel:dword src0_sel:dword src1_sel:dword
+ //! Literal applied on wrong instruction format: v1: %_ = v_mul_f32 %vgpr0, 0x1234 dst_sel:dword src0_sel:dword src1_sel:dword
+ //! Wrong source position for Literal argument: v1: %_ = v_mul_f32 %vgpr0, 0x1234 dst_sel:dword src0_sel:dword src1_sel:dword
bld.vop2_sdwa(aco_opcode::v_mul_f32, bld.def(v1), Operand::c32(0x1234u), inputs[1]);
bld.vop2_sdwa(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], Operand::c32(0x1234u));
bld.vopc_sdwa(aco_opcode::v_cmp_gt_f32, bld.def(bld.lm, vcc), inputs[0], inputs[1]);
- //~gfx8! SDWA+VOPC definition must be fixed to vcc on GFX8: s2: %_ = v_cmp_lt_f32 %vgpr0, %vgpr1
+ //~gfx8! SDWA+VOPC definition must be fixed to vcc on GFX8: s2: %_ = v_cmp_lt_f32 %vgpr0, %vgpr1 src0_sel:dword src1_sel:dword
bld.vopc_sdwa(aco_opcode::v_cmp_lt_f32, bld.def(bld.lm), inputs[0], inputs[1]);
- //~gfx(9|10)! SDWA VOPC clamp only supported on GFX8: s2: %_:vcc = v_cmp_eq_f32 %vgpr0, %vgpr1 clamp
+ //~gfx(9|10)! SDWA VOPC clamp only supported on GFX8: s2: %_:vcc = v_cmp_eq_f32 %vgpr0, %vgpr1 clamp src0_sel:dword src1_sel:dword
bld.vopc_sdwa(aco_opcode::v_cmp_eq_f32, bld.def(bld.lm, vcc), inputs[0], inputs[1]).instr->sdwa().clamp = true;
//! Validation failed
continue;
//>> Validation results:
- //~gfx8! SDWA omod only supported on GFX9+: v1: %_ = v_mul_f32 %vgpr0, %vgpr1 *2
+ //~gfx8! SDWA omod only supported on GFX9+: v1: %_ = v_mul_f32 %vgpr0, %vgpr1 *2 dst_sel:dword src0_sel:dword src1_sel:dword
//~gfx8! Validation failed
//~gfx(9|10)! Validation passed
bld.vop2_sdwa(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]).instr->sdwa().omod = 1;
continue;
//>> Validation results:
- //! 3rd operand must be fixed to vcc with SDWA: v1: %_ = v_cndmask_b32 %vgpr0, %vgpr1, %_
+ //! 3rd operand must be fixed to vcc with SDWA: v1: %_ = v_cndmask_b32 %vgpr0, %vgpr1, %_ dst_sel:dword src0_sel:dword src1_sel:dword
bld.vop2_sdwa(aco_opcode::v_cndmask_b32, bld.def(v1), inputs[0], inputs[1], inputs[2]);
bld.vop2_sdwa(aco_opcode::v_cndmask_b32, bld.def(v1), inputs[0], inputs[1], bld.vcc(inputs[2]));
- //! 2nd definition must be fixed to vcc with SDWA: v1: %_, s2: %_ = v_add_co_u32 %vgpr0, %vgpr1
+ //! 2nd definition must be fixed to vcc with SDWA: v1: %_, s2: %_ = v_add_co_u32 %vgpr0, %vgpr1 dst_sel:dword src0_sel:dword src1_sel:dword
bld.vop2_sdwa(aco_opcode::v_add_co_u32, bld.def(v1), bld.def(bld.lm), inputs[0], inputs[1]);
bld.vop2_sdwa(aco_opcode::v_add_co_u32, bld.def(v1), bld.def(bld.lm, vcc), inputs[0], inputs[1]);
if (!setup_cs("v1 v1 s1 s1", (chip_class)i, CHIP_UNKNOWN, is_signed ? "_signed" : "_unsigned"))
continue;
- //; funcs['b'] = lambda bits: ('sext(%%b)[%s]' if variant.endswith('_signed') else '%%b[%s]') % bits
-
- //; def standard_test(index, offset, size):
- //; res = 'v1: %%res%d = v_mul_f32 %%a, @b(%d:%d)\n' % (index, offset % 32, offset % 32 + size % 32 - 1)
- //; res += 'p_unit_test %d, %%res%d' % (index, index)
+ //; def standard_test(index, sel):
+ //; res = 'v1: %%res%s = v_mul_f32 %%a, %%b dst_sel:dword src0_sel:dword src1_sel:%c%s\n' % (index, 's' if variant.endswith('_signed') else 'u', sel)
+ //; res += 'p_unit_test %s, %%res%s' % (index, index)
//; return res
- //; funcs['standard_test'] = lambda a: standard_test(*(int(v) for v in a.split(',')))
+ //; funcs['standard_test'] = lambda a: standard_test(*(v for v in a.split(',')))
aco_opcode ext = aco_opcode::p_extract;
aco_opcode ins = aco_opcode::p_insert;
{
- //~gfx[^7].*! @standard_test(0, 0, 8)
+ //~gfx[^7].*! @standard_test(0,byte0)
Temp bfe_byte0_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::zero(), Operand::c32(8u),
Operand::c32(is_signed));
writeout(0, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], bfe_byte0_b));
- //~gfx[^7].*! @standard_test(1, 8, 8)
+ //~gfx[^7].*! @standard_test(1,byte1)
Temp bfe_byte1_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::c32(1u), Operand::c32(8u),
Operand::c32(is_signed));
writeout(1, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], bfe_byte1_b));
- //~gfx[^7].*! @standard_test(2, 16, 8)
+ //~gfx[^7].*! @standard_test(2,byte2)
Temp bfe_byte2_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::c32(2u), Operand::c32(8u),
Operand::c32(is_signed));
writeout(2, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], bfe_byte2_b));
- //~gfx[^7].*! @standard_test(3, 24, 8)
+ //~gfx[^7].*! @standard_test(3,byte3)
Temp bfe_byte3_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::c32(3u), Operand::c32(8u),
Operand::c32(is_signed));
writeout(3, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], bfe_byte3_b));
- //~gfx[^7].*! @standard_test(4, 0, 16)
+ //~gfx[^7].*! @standard_test(4,word0)
Temp bfe_word0_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::zero(), Operand::c32(16u),
Operand::c32(is_signed));
writeout(4, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], bfe_word0_b));
- //~gfx[^7].*! @standard_test(5, 16, 16)
+ //~gfx[^7].*! @standard_test(5,word1)
Temp bfe_word1_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::c32(1u),
Operand::c32(16u), Operand::c32(is_signed));
writeout(5, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], bfe_word1_b));
- //~gfx[^7]_unsigned! @standard_test(6, 0, 8)
+ //~gfx[^7]_unsigned! @standard_test(6,byte0)
Temp bfi_byte0_b = bld.pseudo(ins, bld.def(v1), inputs[1], Operand::zero(), Operand::c32(8u));
writeout(6, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], bfi_byte0_b));
- //~gfx[^7]_unsigned! @standard_test(7, 0, 16)
+ //~gfx[^7]_unsigned! @standard_test(7,word0)
Temp bfi_word0_b =
bld.pseudo(ins, bld.def(v1), inputs[1], Operand::zero(), Operand::c32(16u));
writeout(7, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], bfi_word0_b));
/* v_cvt_f32_ubyte[0-3] can be used instead of v_cvt_f32_u32+sdwa */
//~gfx7_signed! v1: %bfe_byte0_b = p_extract %b, 0, 8, 1
//~gfx7_signed! v1: %res9 = v_cvt_f32_u32 %bfe_byte0_b
- //~gfx[^7]+_signed! v1: %res9 = v_cvt_f32_u32 @b(0:7)
+ //~gfx[^7]+_signed! v1: %res9 = v_cvt_f32_u32 %b dst_sel:dword src0_sel:sbyte0
//~gfx\d+_unsigned! v1: %res9 = v_cvt_f32_ubyte0 %b
//! p_unit_test 9, %res9
Temp bfe_byte0_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::zero(), Operand::c32(8u),
//~gfx7_signed! v1: %bfe_byte1_b = p_extract %b, 1, 8, 1
//~gfx7_signed! v1: %res10 = v_cvt_f32_u32 %bfe_byte1_b
- //~gfx[^7]+_signed! v1: %res10 = v_cvt_f32_u32 @b(8:15)
+ //~gfx[^7]+_signed! v1: %res10 = v_cvt_f32_u32 %b dst_sel:dword src0_sel:sbyte1
//~gfx\d+_unsigned! v1: %res10 = v_cvt_f32_ubyte1 %b
//! p_unit_test 10, %res10
Temp bfe_byte1_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::c32(1u), Operand::c32(8u),
//~gfx7_signed! v1: %bfe_byte2_b = p_extract %b, 2, 8, 1
//~gfx7_signed! v1: %res11 = v_cvt_f32_u32 %bfe_byte2_b
- //~gfx[^7]+_signed! v1: %res11 = v_cvt_f32_u32 @b(16:23)
+ //~gfx[^7]+_signed! v1: %res11 = v_cvt_f32_u32 %b dst_sel:dword src0_sel:sbyte2
//~gfx\d+_unsigned! v1: %res11 = v_cvt_f32_ubyte2 %b
//! p_unit_test 11, %res11
Temp bfe_byte2_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::c32(2u), Operand::c32(8u),
//~gfx7_signed! v1: %bfe_byte3_b = p_extract %b, 3, 8, 1
//~gfx7_signed! v1: %res12 = v_cvt_f32_u32 %bfe_byte3_b
- //~gfx[^7]+_signed! v1: %res12 = v_cvt_f32_u32 @b(24:31)
+ //~gfx[^7]+_signed! v1: %res12 = v_cvt_f32_u32 %b dst_sel:dword src0_sel:sbyte3
//~gfx\d+_unsigned! v1: %res12 = v_cvt_f32_ubyte3 %b
//! p_unit_test 12, %res12
Temp bfe_byte3_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::c32(3u), Operand::c32(8u),
aco_opcode ext = aco_opcode::p_extract;
- //! v1: %res0 = v_mul_f32 %a, -%b[0:7]
+ //! v1: %res0 = v_mul_f32 %a, -%b dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 0, %res0
Temp byte0 = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::zero(), Operand::c32(8u),
Operand::zero());
writeout(0, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], neg_byte0));
//~gfx8! v1: %neg = v_mul_f32 -1.0, %b
- //~gfx8! v1: %res1 = v_mul_f32 %a, %neg[0:7]
- //~gfx(9|10)! v1: %neg_byte0 = v_mul_f32 -1.0, %b dst_sel:ubyte0
+ //~gfx8! v1: %res1 = v_mul_f32 %a, %neg dst_sel:dword src0_sel:dword src1_sel:ubyte0
+ //~gfx(9|10)! v1: %neg_byte0 = v_mul_f32 -1.0, %b dst_sel:ubyte0 src0_sel:dword src1_sel:dword
//~gfx(9|10)! v1: %res1 = v_mul_f32 %a, %neg_byte0
//! p_unit_test 1, %res1
Temp neg = fneg(inputs[1]);
bld.pseudo(ext, bld.def(v1), neg, Operand::zero(), Operand::c32(8u), Operand::zero());
writeout(1, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], byte0_neg));
- //! v1: %res2 = v_mul_f32 %a, |%b[0:7]|
+ //! v1: %res2 = v_mul_f32 %a, |%b| dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 2, %res2
Temp abs_byte0 = fabs(byte0);
writeout(2, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], abs_byte0));
//! v1: %abs = v_mul_f32 1.0, |%b|
- //! v1: %res3 = v_mul_f32 %a, %abs[0:7]
+ //! v1: %res3 = v_mul_f32 %a, %abs dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 3, %res3
Temp abs = fabs(inputs[1]);
Temp byte0_abs =
bld.pseudo(ext, bld.def(v1), abs, Operand::zero(), Operand::c32(8u), Operand::zero());
writeout(3, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], byte0_abs));
- //! v1: %res4 = v_mul_f32 %1, -|%2[0:7]|
+ //! v1: %res4 = v_mul_f32 %1, -|%2| dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 4, %res4
Temp neg_abs_byte0 = fneg(abs_byte0);
writeout(4, bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], neg_abs_byte0));
//~gfx8! v1: %neg_abs = v_mul_f32 -1.0, %abs
- //~gfx8! v1: %res5 = v_mul_f32 %a, %neg_abs[0:7]
- //~gfx(9|10)! v1: %neg_abs_byte0 = v_mul_f32 -1.0, %abs dst_sel:ubyte0
+ //~gfx8! v1: %res5 = v_mul_f32 %a, %neg_abs dst_sel:dword src0_sel:dword src1_sel:ubyte0
+ //~gfx(9|10)! v1: %neg_abs_byte0 = v_mul_f32 -1.0, %abs dst_sel:ubyte0 src0_sel:dword src1_sel:dword
//~gfx(9|10)! v1: %res5 = v_mul_f32 %a, %neg_abs_byte0
//! p_unit_test 5, %res5
Temp neg_abs = fneg(abs);
//~gfx8! v1: %byte0_b = p_extract %b, 0, 8, 0
//~gfx8! v1: %res1 = v_mul_f32 %c, %byte0_b
- //~gfx(9|10)! v1: %res1 = v_mul_f32 %c, %b[0:7]
+ //~gfx(9|10)! v1: %res1 = v_mul_f32 %c, %b dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 1, %res1
Temp byte0_b = bld.pseudo(ext, bld.def(v1), inputs[1], Operand::zero(), Operand::c32(8u),
Operand::zero());
//~gfx8! v1: %byte0_c = p_extract %c, 0, 8, 0
//~gfx8! v1: %res2 = v_mul_f32 %a, %byte0_c
- //~gfx(9|10)! v1: %res2 = v_mul_f32 %a, %c[0:7]
+ //~gfx(9|10)! v1: %res2 = v_mul_f32 %a, %c dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 2, %res2
Temp byte0_c = bld.pseudo(ext, bld.def(v1), inputs[2], Operand::zero(), Operand::c32(8u),
Operand::zero());
//~gfx8! v1: %byte0_c_2 = p_extract %c, 0, 8, 0
//~gfx8! v1: %res3 = v_mul_f32 %c, %byte0_c_2
- //~gfx(9|10)! v1: %res3 = v_mul_f32 %c, %c[0:7]
+ //~gfx(9|10)! v1: %res3 = v_mul_f32 %c, %c dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 3, %res3
byte0_c = bld.pseudo(ext, bld.def(v1), inputs[2], Operand::zero(), Operand::c32(8u),
Operand::zero());
//~gfx(8|9)! v1: %byte0_c_3 = p_extract %c, 0, 8, 0
//~gfx(8|9)! v1: %res4 = v_mul_f32 %d, %byte0_c_3
- //~gfx10! v1: %res4 = v_mul_f32 %d, %c[0:7]
+ //~gfx10! v1: %res4 = v_mul_f32 %d, %c dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 4, %res4
byte0_c = bld.pseudo(ext, bld.def(v1), inputs[2], Operand::zero(), Operand::c32(8u),
Operand::zero());
if (!setup_cs("v1 v1 s1 s1", (chip_class)i))
continue;
- //! v1: %res0 = v_mul_f32 -|%a|, %b[0:7]
+ //! v1: %res0 = v_mul_f32 -|%a|, %b dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 0, %res0
Temp byte0_b = bld.pseudo(aco_opcode::p_extract, bld.def(v1), inputs[1], Operand::zero(),
Operand::c32(8u), Operand::zero());
//~gfx8! v1: %byte0_b_0 = p_extract %b, 0, 8, 0
//~gfx8! v1: %res1 = v_mul_f32 %a, %byte0_b_0 *4
- //~gfx(9|10)! v1: %res1 = v_mul_f32 %a, %b[0:7] *4
+ //~gfx(9|10)! v1: %res1 = v_mul_f32 %a, %b *4 dst_sel:dword src0_sel:dword src1_sel:ubyte0
//! p_unit_test 1, %res1
byte0_b = bld.pseudo(aco_opcode::p_extract, bld.def(v1), inputs[1], Operand::zero(),
Operand::c32(8u), Operand::zero());
//~gfx8! v1: %byte0_b_1 = p_extract %b, 0, 8, 0
//~gfx8! v1: %res2 = v_mul_f32 %byte0_b_1, %c
- //~gfx(9|10)! v1: %res2 = v_mul_f32 %b[0:7], %c
+ //~gfx(9|10)! v1: %res2 = v_mul_f32 %b, %c dst_sel:dword src0_sel:ubyte0 src1_sel:dword
//! p_unit_test 2, %res2
byte0_b = bld.pseudo(aco_opcode::p_extract, bld.def(v1), inputs[1], Operand::zero(),
Operand::c32(8u), Operand::zero());
aco_opcode ext = aco_opcode::p_extract;
aco_opcode ins = aco_opcode::p_insert;
- //~gfx[^7]! v1: %res0 = v_mul_f32 %a, %b dst_sel:ubyte0
+ //~gfx[^7]! v1: %res0 = v_mul_f32 %a, %b dst_sel:ubyte0 src0_sel:dword src1_sel:dword
//~gfx[^7]! p_unit_test 0, %res0
Temp val = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
writeout(0, bld.pseudo(ins, bld.def(v1), val, Operand::zero(), Operand::c32(8u)));
- //~gfx[^7]! v1: %res1 = v_mul_f32 %a, %b dst_sel:ubyte1
+ //~gfx[^7]! v1: %res1 = v_mul_f32 %a, %b dst_sel:ubyte1 src0_sel:dword src1_sel:dword
//~gfx[^7]! p_unit_test 1, %res1
val = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
writeout(1, bld.pseudo(ins, bld.def(v1), val, Operand::c32(1u), Operand::c32(8u)));
- //~gfx[^7]! v1: %res2 = v_mul_f32 %a, %b dst_sel:ubyte2
+ //~gfx[^7]! v1: %res2 = v_mul_f32 %a, %b dst_sel:ubyte2 src0_sel:dword src1_sel:dword
//~gfx[^7]! p_unit_test 2, %res2
val = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
writeout(2, bld.pseudo(ins, bld.def(v1), val, Operand::c32(2u), Operand::c32(8u)));
- //~gfx[^7]! v1: %res3 = v_mul_f32 %a, %b dst_sel:ubyte3
+ //~gfx[^7]! v1: %res3 = v_mul_f32 %a, %b dst_sel:ubyte3 src0_sel:dword src1_sel:dword
//~gfx[^7]! p_unit_test 3, %res3
val = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
writeout(3, bld.pseudo(ins, bld.def(v1), val, Operand::c32(3u), Operand::c32(8u)));
- //~gfx[^7]! v1: %res4 = v_mul_f32 %a, %b dst_sel:uword0
+ //~gfx[^7]! v1: %res4 = v_mul_f32 %a, %b dst_sel:uword0 src0_sel:dword src1_sel:dword
//~gfx[^7]! p_unit_test 4, %res4
val = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
writeout(4, bld.pseudo(ins, bld.def(v1), val, Operand::zero(), Operand::c32(16u)));
- //~gfx[^7]! v1: %res5 = v_mul_f32 %a, %b dst_sel:uword1
+ //~gfx[^7]! v1: %res5 = v_mul_f32 %a, %b dst_sel:uword1 src0_sel:dword src1_sel:dword
//~gfx[^7]! p_unit_test 5, %res5
val = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
writeout(5, bld.pseudo(ins, bld.def(v1), val, Operand::c32(1u), Operand::c32(16u)));
- //~gfx[^7]! v1: %res6 = v_mul_f32 %a, %b dst_sel:ubyte0
+ //~gfx[^7]! v1: %res6 = v_mul_f32 %a, %b dst_sel:ubyte0 src0_sel:dword src1_sel:dword
//~gfx[^7]! p_unit_test 6, %res6
val = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
writeout(
6, bld.pseudo(ext, bld.def(v1), val, Operand::zero(), Operand::c32(8u), Operand::zero()));
- //~gfx[^7]! v1: %res7 = v_mul_f32 %a, %b dst_sel:uword0
+ //~gfx[^7]! v1: %res7 = v_mul_f32 %a, %b dst_sel:uword0 src0_sel:dword src1_sel:dword
//~gfx[^7]! p_unit_test 7, %res7
val = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), inputs[0], inputs[1]);
writeout(
//~gfx8! v1: %tmp0 = v_rcp_f32 %a *2
//~gfx8! v1: %res0 = p_insert %tmp0, 0, 8
- //~gfx9! v1: %res0 = v_rcp_f32 %a *2 dst_sel:ubyte0
+ //~gfx9! v1: %res0 = v_rcp_f32 %a *2 dst_sel:ubyte0 src0_sel:dword
//! p_unit_test 0, %res0
Temp val = bld.vop1(aco_opcode::v_rcp_f32, bld.def(v1), inputs[0]);
val = bld.vop2_e64(aco_opcode::v_mul_f32, bld.def(v1), val, Operand::c32(0x40000000u));
writeout(0, bld.pseudo(ins, bld.def(v1), val, Operand::zero(), Operand::c32(8u)));
- //! v1: %res1 = v_rcp_f32 %a clamp dst_sel:ubyte0
+ //! v1: %res1 = v_rcp_f32 %a clamp dst_sel:ubyte0 src0_sel:dword
//! p_unit_test 1, %res1
val = bld.vop1(aco_opcode::v_rcp_f32, bld.def(v1), inputs[0]);
val = bld.vop3(aco_opcode::v_med3_f32, bld.def(v1), val, Operand::zero(),
Operand::c32(0x3f800000u));
writeout(1, bld.pseudo(ins, bld.def(v1), val, Operand::zero(), Operand::c32(8u)));
- //! v1: %tmp2 = v_rcp_f32 %a dst_sel:ubyte0
+ //! v1: %tmp2 = v_rcp_f32 %a dst_sel:ubyte0 src0_sel:dword
//! v1: %res2 = v_mul_f32 %tmp2, 2.0
//! p_unit_test 2, %res2
val = bld.vop1(aco_opcode::v_rcp_f32, bld.def(v1), inputs[0]);
val = bld.vop2_e64(aco_opcode::v_mul_f32, bld.def(v1), val, Operand::c32(0x40000000u));
writeout(2, val);
- //! v1: %tmp3 = v_rcp_f32 %a dst_sel:ubyte0
+ //! v1: %tmp3 = v_rcp_f32 %a dst_sel:ubyte0 src0_sel:dword
//! v1: %res3 = v_med3_f32 %tmp3, 0, 1.0
//! p_unit_test 3, %res3
val = bld.vop1(aco_opcode::v_rcp_f32, bld.def(v1), inputs[0]);
//~gfx8! v1: %tmp4 = v_rcp_f32 %a *2 clamp
//~gfx8! v1: %res4 = p_insert %tmp4, 0, 8
- //~gfx9! v1: %res4 = v_rcp_f32 %a *2 clamp dst_sel:ubyte0
+ //~gfx9! v1: %res4 = v_rcp_f32 %a *2 clamp dst_sel:ubyte0 src0_sel:dword
//! p_unit_test 4, %res4
val = bld.vop1(aco_opcode::v_rcp_f32, bld.def(v1), inputs[0]);
val = bld.vop2_e64(aco_opcode::v_mul_f32, bld.def(v1), val, Operand::c32(0x40000000u));
//~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]
- //~gfx[89]! 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_sel:uword1 dst_preserve src0_sel:uword1
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(1u));
bld.pseudo(aco_opcode::p_parallelcopy,
Definition(v0_lo, v1), Definition(v1_lo, v2b),
Operand(v1_lo, v1), Operand(v0_lo, v2b));
//~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
+ //~gfx[89]! v2b: %0:v[0][16:32] = v_mov_b32 %0:v[1][16:32] dst_sel:uword1 dst_preserve src0_sel:uword1
+ //~gfx[89]! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][0:16] dst_sel:uword1 dst_preserve src0_sel:uword0
+ //~gfx[89]! v2b: %0:v[1][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_sel:uword0 dst_preserve src0_sel:uword0 src1_sel:uword0
+ //~gfx[89]! v2b: %0:v[0][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_sel:uword0 dst_preserve src0_sel:uword0 src1_sel:uword0
+ //~gfx[89]! v2b: %0:v[1][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_sel:uword0 dst_preserve src0_sel:uword0 src1_sel:uword0
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(2u));
bld.pseudo(aco_opcode::p_parallelcopy,
Definition(v0_lo, v1), Definition(v1_lo, v2b), Definition(v1_hi, v2b),
//~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]
- //~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
+ //~gfx[89]! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][0:16] dst_sel:uword0 dst_preserve src0_sel:uword0
+ //~gfx[89]! v1b: %0:v[1][16:24] = v_mov_b32 %0:v[0][16:24] dst_sel:ubyte2 dst_preserve src0_sel:ubyte2
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(3u));
bld.pseudo(aco_opcode::p_parallelcopy,
Definition(v0_lo, v1), Definition(v1_b3, v1b),
//~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]
- //~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
+ //~gfx[89]! v1b: %0:v[1][8:16] = v_mov_b32 %0:v[0][8:16] dst_sel:ubyte1 dst_preserve src0_sel:ubyte1
+ //~gfx[89]! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][16:32] dst_sel:uword1 dst_preserve src0_sel:uword1
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(4u));
bld.pseudo(aco_opcode::p_parallelcopy,
Definition(v0_lo, v1), Definition(v1_lo, v1b),
//~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]
- //~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
+ //~gfx[89]! v1b: %0:v[0][8:16] = v_mov_b32 %0:v[1][8:16] dst_sel:ubyte1 dst_preserve src0_sel:ubyte1
+ //~gfx[89]! v1b: %0:v[0][24:32] = v_mov_b32 %0:v[1][24:32] dst_sel:ubyte3 dst_preserve src0_sel:ubyte3
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(5u));
bld.pseudo(aco_opcode::p_parallelcopy,
Definition(v0_lo, v1b), Definition(v0_hi, v1b), Definition(v1_lo, v1),
//~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]
- //~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
+ //~gfx[89]! v1b: %0:v[1][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_sel:ubyte3 dst_preserve src0_sel:ubyte3 src1_sel:ubyte3
+ //~gfx[89]! v1b: %0:v[0][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_sel:ubyte3 dst_preserve src0_sel:ubyte3 src1_sel:ubyte3
+ //~gfx[89]! v1b: %0:v[1][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_sel:ubyte3 dst_preserve src0_sel:ubyte3 src1_sel:ubyte3
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(8u));
bld.pseudo(aco_opcode::p_parallelcopy,
Definition(v0_lo, v3b), Definition(v1_lo, v3b),
//~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]
- //~gfx[89]! 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_sel:ubyte3 dst_preserve src0_sel:ubyte3
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(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));
//~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
+ //~gfx[89]! v1b: %0:v[1][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_sel:ubyte1 dst_preserve src0_sel:ubyte1 src1_sel:ubyte1
+ //~gfx[89]! v1b: %0:v[0][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_sel:ubyte1 dst_preserve src0_sel:ubyte1 src1_sel:ubyte1
+ //~gfx[89]! v1b: %0:v[1][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_sel:ubyte1 dst_preserve src0_sel:ubyte1 src1_sel:ubyte1
+ //~gfx[89]! v1b: %0:v[1][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_sel:ubyte2 dst_preserve src0_sel:ubyte2 src1_sel:ubyte2
+ //~gfx[89]! v1b: %0:v[0][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_sel:ubyte2 dst_preserve src0_sel:ubyte2 src1_sel:ubyte2
+ //~gfx[89]! v1b: %0:v[1][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_sel:ubyte2 dst_preserve src0_sel:ubyte2 src1_sel:ubyte2
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(10u));
bld.pseudo(aco_opcode::p_parallelcopy,
Definition(v0_b1, v2b), Definition(v1_b1, v2b),
Operand(v1_b1, v2b), Operand(v0_b1, v2b));
//~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]! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][16:32] dst_sel:uword0 dst_preserve src0_sel:uword1
//~gfx[89]! v1: %0:v[0] = v_mov_b32 42
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(11u));
bld.pseudo(aco_opcode::p_parallelcopy, Definition(v0_lo, v1), Definition(v1_lo, v2b),
/* 16-bit copy */
//! p_unit_test 6
- //! v2b: %_:v[0][0:16] = v_add_f16 0.5, 0 dst_preserve
+ //! v2b: %_:v[0][0:16] = v_add_f16 0.5, 0 dst_sel:uword0 dst_preserve src0_sel:uword0 src1_sel:dword
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(6u));
bld.pseudo(aco_opcode::p_parallelcopy, Definition(v0_lo, v2b), Operand::c16(0x3800));
bld.pseudo(aco_opcode::p_parallelcopy, Definition(v0_hi, v2b), Operand::c16(0x4205));
//! p_unit_test 9
- //! v1b: %_:v[0][8:16] = v_mov_b32 0 dst_preserve
- //! v1b: %_:v[0][16:24] = v_mov_b32 56 dst_preserve
+ //! v1b: %_:v[0][8:16] = v_mov_b32 0 dst_sel:ubyte1 dst_preserve src0_sel:dword
+ //! v1b: %_:v[0][16:24] = v_mov_b32 56 dst_sel:ubyte2 dst_preserve src0_sel:dword
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(9u));
bld.pseudo(aco_opcode::p_parallelcopy, Definition(v0_b1, v2b), Operand::c16(0x3800));
//! p_unit_test 10
- //! v1b: %_:v[0][8:16] = v_mov_b32 5 dst_preserve
- //! v1b: %_:v[0][16:24] = v_mul_u32_u24 2, 33 dst_preserve
+ //! v1b: %_:v[0][8:16] = v_mov_b32 5 dst_sel:ubyte1 dst_preserve src0_sel:dword
+ //! v1b: %_:v[0][16:24] = v_mul_u32_u24 2, 33 dst_sel:ubyte2 dst_preserve src0_sel:dword src1_sel:dword
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(10u));
bld.pseudo(aco_opcode::p_parallelcopy, Definition(v0_b1, v2b), Operand::c16(0x4205));
/* 8-bit copy */
//! p_unit_test 11
- //! v1b: %_:v[0][0:8] = v_mul_u32_u24 2, 33 dst_preserve
+ //! v1b: %_:v[0][0:8] = v_mul_u32_u24 2, 33 dst_sel:ubyte0 dst_preserve src0_sel:dword src1_sel:dword
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(11u));
bld.pseudo(aco_opcode::p_parallelcopy, Definition(v0_lo, v1b), Operand::c8(0x42));
//; funcs['v_shr'] = lambda _: 'v_ashrrev_i32' if variant.endswith('_signed') else 'v_lshrrev_b32'
//; funcs['s_bfe'] = lambda _: 's_bfe_i32' if variant.endswith('_signed') else 's_bfe_u32'
//; funcs['s_shr'] = lambda _: 's_ashr_i32' if variant.endswith('_signed') else 's_lshr_b32'
- //; funcs['sel'] = lambda bits: ('sext(%%_:v[1])[%s]' if variant.endswith('_signed') else '%%_:v[1][%s]') % bits
+ //; funcs['byte'] = lambda n: '%cbyte%s' % ('s' if variant.endswith('_signed') else 'u', n)
//>> p_unit_test 0
bld.pseudo(aco_opcode::p_unit_test, Operand::zero());
//>> p_unit_test 4
bld.pseudo(aco_opcode::p_unit_test, Operand::c32(4u));
//~gfx7.*! v2b: %_:v[0][0:16] = @v_bfe %_:v[1][0:16], 0, 8
- //~gfx[^7].*! v2b: %_:v[0][0:16] = v_mov_b32 @sel(0:7) dst_preserve
+ //~gfx[^7].*! v2b: %_:v[0][0:16] = v_mov_b32 %_:v[1] dst_sel:uword0 dst_preserve src0_sel:@byte(0)
EXT(0, 0)
- //~gfx[^7].*! v2b: %_:v[0][0:16] = v_mov_b32 @sel(16:23) dst_preserve
+ //~gfx[^7].*! v2b: %_:v[0][0:16] = v_mov_b32 %_:v[1] dst_sel:uword0 dst_preserve src0_sel:@byte(2)
if (i != GFX7)
EXT(0, 2)
//~gfx7.*! v2b: %_:v[0][0:16] = @v_bfe %_:v[1][0:16], 8, 8
- //~gfx[^7].*! v2b: %_:v[0][0:16] = v_mov_b32 @sel(8:15) dst_preserve
+ //~gfx[^7].*! v2b: %_:v[0][0:16] = v_mov_b32 %_:v[1] dst_sel:uword0 dst_preserve src0_sel:@byte(1)
EXT(1, 0)
- //~gfx[^7].*! v2b: %_:v[0][0:16] = v_mov_b32 @sel(24:31) dst_preserve
+ //~gfx[^7].*! v2b: %_:v[0][0:16] = v_mov_b32 %_:v[1] dst_sel:uword0 dst_preserve src0_sel:@byte(3)
if (i != GFX7)
EXT(1, 2)
INS(0, 8)
//~gfx7! v1: %0:v[0] = v_bfe_u32 %0:v[1], 0, 8
//~gfx7! v1: %0:v[0] = v_lshlrev_b32 8, %0:v[0]
- //~gfx[^7]! v1: %0:v[0] = v_mov_b32 %0:v[1] dst_sel:ubyte1
+ //~gfx[^7]! v1: %0:v[0] = v_mov_b32 %0:v[1] dst_sel:ubyte1 src0_sel:dword
INS(1, 8)
//~gfx7! v1: %0:v[0] = v_bfe_u32 %0:v[1], 0, 8
//~gfx7! v1: %0:v[0] = v_lshlrev_b32 16, %0:v[0]
- //~gfx[^7]! v1: %0:v[0] = v_mov_b32 %0:v[1] dst_sel:ubyte2
+ //~gfx[^7]! v1: %0:v[0] = v_mov_b32 %0:v[1] dst_sel:ubyte2 src0_sel:dword
INS(2, 8)
//! v1: %0:v[0] = v_lshlrev_b32 24, %0:v[1]
INS(3, 8)
//>> 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]! v2b: %0:v[0][0:16] = v_lshlrev_b32 0, %0:v[1][0:7] dst_preserve
+ //~gfx[^7]! v2b: %0:v[0][0:16] = v_lshlrev_b32 0, %0:v[1] dst_sel:uword0 dst_preserve src0_sel:dword src1_sel:ubyte0
INS(0, 0)
- //~gfx[^7]! v2b: %0:v[0][16:32] = v_lshlrev_b32 0, %0:v[1][0:7] dst_preserve
+ //~gfx[^7]! v2b: %0:v[0][16:32] = v_lshlrev_b32 0, %0:v[1] dst_sel:uword1 dst_preserve src0_sel:dword src1_sel:ubyte0
if (i != GFX7)
INS(0, 2)
//~gfx7! v2b: %_:v[0][0:16] = v_lshlrev_b32 8, %_:v[1][0:16]
- //~gfx[^7]! v2b: %0:v[0][0:16] = v_lshlrev_b32 8, %0:v[1][0:7] dst_preserve
+ //~gfx[^7]! v2b: %0:v[0][0:16] = v_lshlrev_b32 8, %0:v[1] dst_sel:uword0 dst_preserve src0_sel:dword src1_sel:ubyte0
INS(1, 0)
- //~gfx[^7]! v2b: %0:v[0][16:32] = v_lshlrev_b32 8, %0:v[1][0:7] dst_preserve
+ //~gfx[^7]! v2b: %0:v[0][16:32] = v_lshlrev_b32 8, %0:v[1] dst_sel:uword1 dst_preserve src0_sel:dword src1_sel:ubyte0
if (i != GFX7)
INS(1, 2)