} // End hasSideEffects = 1, Defs = [EXEC]
defm V_CMP_F_F64 : VOPC_64 <0x00000020, "V_CMP_F_F64">;
-defm V_CMP_LT_F64 : VOPC_64 <0x00000021, "V_CMP_LT_F64">;
-defm V_CMP_EQ_F64 : VOPC_64 <0x00000022, "V_CMP_EQ_F64">;
-defm V_CMP_LE_F64 : VOPC_64 <0x00000023, "V_CMP_LE_F64">;
-defm V_CMP_GT_F64 : VOPC_64 <0x00000024, "V_CMP_GT_F64">;
+defm V_CMP_LT_F64 : VOPC_64 <0x00000021, "V_CMP_LT_F64", f64, COND_LT>;
+defm V_CMP_EQ_F64 : VOPC_64 <0x00000022, "V_CMP_EQ_F64", f64, COND_EQ>;
+defm V_CMP_LE_F64 : VOPC_64 <0x00000023, "V_CMP_LE_F64", f64, COND_LE>;
+defm V_CMP_GT_F64 : VOPC_64 <0x00000024, "V_CMP_GT_F64", f64, COND_GT>;
defm V_CMP_LG_F64 : VOPC_64 <0x00000025, "V_CMP_LG_F64">;
-defm V_CMP_GE_F64 : VOPC_64 <0x00000026, "V_CMP_GE_F64">;
+defm V_CMP_GE_F64 : VOPC_64 <0x00000026, "V_CMP_GE_F64", f64, COND_GE>;
defm V_CMP_O_F64 : VOPC_64 <0x00000027, "V_CMP_O_F64">;
defm V_CMP_U_F64 : VOPC_64 <0x00000028, "V_CMP_U_F64">;
defm V_CMP_NGE_F64 : VOPC_64 <0x00000029, "V_CMP_NGE_F64">;
defm V_CMP_NLG_F64 : VOPC_64 <0x0000002a, "V_CMP_NLG_F64">;
defm V_CMP_NGT_F64 : VOPC_64 <0x0000002b, "V_CMP_NGT_F64">;
defm V_CMP_NLE_F64 : VOPC_64 <0x0000002c, "V_CMP_NLE_F64">;
-defm V_CMP_NEQ_F64 : VOPC_64 <0x0000002d, "V_CMP_NEQ_F64">;
+defm V_CMP_NEQ_F64 : VOPC_64 <0x0000002d, "V_CMP_NEQ_F64", f64, COND_NE>;
defm V_CMP_NLT_F64 : VOPC_64 <0x0000002e, "V_CMP_NLT_F64">;
defm V_CMP_TRU_F64 : VOPC_64 <0x0000002f, "V_CMP_TRU_F64">;
//def BUFFER_STORE_FORMAT_XYZ : MUBUF_ <0x00000006, "BUFFER_STORE_FORMAT_XYZ", []>;
//def BUFFER_STORE_FORMAT_XYZW : MUBUF_ <0x00000007, "BUFFER_STORE_FORMAT_XYZW", []>;
defm BUFFER_LOAD_UBYTE : MUBUF_Load_Helper <0x00000008, "BUFFER_LOAD_UBYTE", VReg_32>;
-//def BUFFER_LOAD_SBYTE : MUBUF_ <0x00000009, "BUFFER_LOAD_SBYTE", []>;
-//def BUFFER_LOAD_USHORT : MUBUF_ <0x0000000a, "BUFFER_LOAD_USHORT", []>;
-//def BUFFER_LOAD_SSHORT : MUBUF_ <0x0000000b, "BUFFER_LOAD_SSHORT", []>;
+defm BUFFER_LOAD_SBYTE : MUBUF_Load_Helper <0x00000009, "BUFFER_LOAD_SBYTE", VReg_32>;
+defm BUFFER_LOAD_USHORT : MUBUF_Load_Helper <0x0000000a, "BUFFER_LOAD_USHORT", VReg_32>;
+defm BUFFER_LOAD_SSHORT : MUBUF_Load_Helper <0x0000000b, "BUFFER_LOAD_SSHORT", VReg_32>;
defm BUFFER_LOAD_DWORD : MUBUF_Load_Helper <0x0000000c, "BUFFER_LOAD_DWORD", VReg_32>;
defm BUFFER_LOAD_DWORDX2 : MUBUF_Load_Helper <0x0000000d, "BUFFER_LOAD_DWORDX2", VReg_64>;
defm BUFFER_LOAD_DWORDX4 : MUBUF_Load_Helper <0x0000000e, "BUFFER_LOAD_DWORDX4", VReg_128>;
} // End neverHasSideEffects = 1, isMoveImm = 1
defm V_READFIRSTLANE_B32 : VOP1_32 <0x00000002, "V_READFIRSTLANE_B32", []>;
-//defm V_CVT_I32_F64 : VOP1_32 <0x00000003, "V_CVT_I32_F64", []>;
-//defm V_CVT_F64_I32 : VOP1_64 <0x00000004, "V_CVT_F64_I32", []>;
+defm V_CVT_I32_F64 : VOP1_32_64 <0x00000003, "V_CVT_I32_F64",
+ [(set i32:$dst, (fp_to_sint f64:$src0))]
+>;
+defm V_CVT_F64_I32 : VOP1_64_32 <0x00000004, "V_CVT_F64_I32",
+ [(set f64:$dst, (sint_to_fp i32:$src0))]
+>;
defm V_CVT_F32_I32 : VOP1_32 <0x00000005, "V_CVT_F32_I32",
[(set f32:$dst, (sint_to_fp i32:$src0))]
>;
//defm V_CVT_RPI_I32_F32 : VOP1_32 <0x0000000c, "V_CVT_RPI_I32_F32", []>;
//defm V_CVT_FLR_I32_F32 : VOP1_32 <0x0000000d, "V_CVT_FLR_I32_F32", []>;
//defm V_CVT_OFF_F32_I4 : VOP1_32 <0x0000000e, "V_CVT_OFF_F32_I4", []>;
-//defm V_CVT_F32_F64 : VOP1_32 <0x0000000f, "V_CVT_F32_F64", []>;
-//defm V_CVT_F64_F32 : VOP1_64 <0x00000010, "V_CVT_F64_F32", []>;
+defm V_CVT_F32_F64 : VOP1_32_64 <0x0000000f, "V_CVT_F32_F64",
+ [(set f32:$dst, (fround f64:$src0))]
+>;
+defm V_CVT_F64_F32 : VOP1_64_32 <0x00000010, "V_CVT_F64_F32",
+ [(set f64:$dst, (fextend f32:$src0))]
+>;
//defm V_CVT_F32_UBYTE0 : VOP1_32 <0x00000011, "V_CVT_F32_UBYTE0", []>;
//defm V_CVT_F32_UBYTE1 : VOP1_32 <0x00000012, "V_CVT_F32_UBYTE1", []>;
//defm V_CVT_F32_UBYTE2 : VOP1_32 <0x00000013, "V_CVT_F32_UBYTE2", []>;
[(set f32:$dst, (int_AMDGPU_rsq f32:$src0))]
>;
defm V_RSQ_F32 : VOP1_32 <0x0000002e, "V_RSQ_F32", []>;
-defm V_RCP_F64 : VOP1_64 <0x0000002f, "V_RCP_F64", []>;
+defm V_RCP_F64 : VOP1_64 <0x0000002f, "V_RCP_F64",
+ [(set f64:$dst, (fdiv FP_ONE, f64:$src0))]
+>;
defm V_RCP_CLAMP_F64 : VOP1_64 <0x00000030, "V_RCP_CLAMP_F64", []>;
defm V_RSQ_F64 : VOP1_64 <0x00000031, "V_RSQ_F64", []>;
defm V_RSQ_CLAMP_F64 : VOP1_64 <0x00000032, "V_RSQ_CLAMP_F64", []>;
-defm V_SQRT_F32 : VOP1_32 <0x00000033, "V_SQRT_F32", []>;
-defm V_SQRT_F64 : VOP1_64 <0x00000034, "V_SQRT_F64", []>;
+defm V_SQRT_F32 : VOP1_32 <0x00000033, "V_SQRT_F32",
+ [(set f32:$dst, (fsqrt f32:$src0))]
+>;
+defm V_SQRT_F64 : VOP1_64 <0x00000034, "V_SQRT_F64",
+ [(set f64:$dst, (fsqrt f64:$src0))]
+>;
defm V_SIN_F32 : VOP1_32 <0x00000035, "V_SIN_F32", []>;
defm V_COS_F32 : VOP1_32 <0x00000036, "V_COS_F32", []>;
defm V_NOT_B32 : VOP1_32 <0x00000037, "V_NOT_B32", []>;
(V_CNDMASK_B32_e64 $src0, $src1, $src2)
>;
+//use two V_CNDMASK_B32_e64 instructions for f64
+def : Pat <
+ (f64 (select i1:$src2, f64:$src1, f64:$src0)),
+ (INSERT_SUBREG (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
+ (V_CNDMASK_B32_e64 (EXTRACT_SUBREG $src0, sub0),
+ (EXTRACT_SUBREG $src1, sub0),
+ $src2), sub0),
+ (V_CNDMASK_B32_e64 (EXTRACT_SUBREG $src0, sub1),
+ (EXTRACT_SUBREG $src1, sub1),
+ $src2), sub1)
+>;
+
defm V_READLANE_B32 : VOP2_32 <0x00000001, "V_READLANE_B32", []>;
defm V_WRITELANE_B32 : VOP2_32 <0x00000002, "V_WRITELANE_B32", []>;
[(set f32:$dst, (fmul f32:$src0, f32:$src1))]
>;
-} // End isCommutable = 1
-//defm V_MUL_I32_I24 : VOP2_32 <0x00000009, "V_MUL_I32_I24", []>;
+defm V_MUL_I32_I24 : VOP2_32 <0x00000009, "V_MUL_I32_I24",
+ [(set i32:$dst, (mul I24:$src0, I24:$src1))]
+>;
//defm V_MUL_HI_I32_I24 : VOP2_32 <0x0000000a, "V_MUL_HI_I32_I24", []>;
-//defm V_MUL_U32_U24 : VOP2_32 <0x0000000b, "V_MUL_U32_U24", []>;
+defm V_MUL_U32_U24 : VOP2_32 <0x0000000b, "V_MUL_U32_U24",
+ [(set i32:$dst, (mul U24:$src0, U24:$src1))]
+>;
//defm V_MUL_HI_U32_U24 : VOP2_32 <0x0000000c, "V_MUL_HI_U32_U24", []>;
-let isCommutable = 1 in {
defm V_MIN_LEGACY_F32 : VOP2_32 <0x0000000d, "V_MIN_LEGACY_F32",
[(set f32:$dst, (AMDGPUfmin f32:$src0, f32:$src1))]
def V_MAD_LEGACY_F32 : VOP3_32 <0x00000140, "V_MAD_LEGACY_F32", []>;
def V_MAD_F32 : VOP3_32 <0x00000141, "V_MAD_F32", []>;
-//def V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24", []>;
-//def V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24", []>;
+def V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24",
+ [(set i32:$dst, (add (mul I24:$src0, I24:$src1), i32:$src2))]
+>;
+def V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24",
+ [(set i32:$dst, (add (mul U24:$src0, U24:$src1), i32:$src2))]
+>;
} // End neverHasSideEffects
def V_CUBEID_F32 : VOP3_32 <0x00000144, "V_CUBEID_F32", []>;
def V_LSHR_B64 : VOP3_64_Shift <0x00000162, "V_LSHR_B64",
[(set i64:$dst, (srl i64:$src0, i32:$src1))]
>;
-def V_ASHR_I64 : VOP3_64_Shift <0x00000163, "V_ASHR_I64", []>;
+def V_ASHR_I64 : VOP3_64_Shift <0x00000163, "V_ASHR_I64",
+ [(set i64:$dst, (sra i64:$src0, i32:$src1))]
+>;
+
+let isCommutable = 1 in {
def V_ADD_F64 : VOP3_64 <0x00000164, "V_ADD_F64", []>;
def V_MUL_F64 : VOP3_64 <0x00000165, "V_MUL_F64", []>;
def V_MIN_F64 : VOP3_64 <0x00000166, "V_MIN_F64", []>;
def V_MAX_F64 : VOP3_64 <0x00000167, "V_MAX_F64", []>;
+
+} // isCommutable = 1
+
+def : Pat <
+ (fadd f64:$src0, f64:$src1),
+ (V_ADD_F64 $src0, $src1, (i64 0))
+>;
+
+def : Pat <
+ (fmul f64:$src0, f64:$src1),
+ (V_MUL_F64 $src0, $src1, (i64 0))
+>;
+
def V_LDEXP_F64 : VOP3_64 <0x00000168, "V_LDEXP_F64", []>;
let isCommutable = 1 in {
} // Uses = [EXEC,VCC,M0], Defs = [EXEC,VCC,M0]
-// This psuedo instruction takes a pointer as input and outputs a resource
-// constant that can be used with the ADDR64 MUBUF instructions.
-
let usesCustomInserter = 1 in {
+// This psuedo instruction takes a pointer as input and outputs a resource
+// constant that can be used with the ADDR64 MUBUF instructions.
def SI_ADDR64_RSRC : InstSI <
(outs SReg_128:$srsrc),
(ins SReg_64:$ptr),
"", []
>;
+def V_SUB_F64 : InstSI <
+ (outs VReg_64:$dst),
+ (ins VReg_64:$src0, VReg_64:$src1),
+ "V_SUB_F64 $dst, $src0, $src1",
+ []
+>;
+
} // end usesCustomInserter
} // end IsCodeGenOnly, isPseudo
$src0, $src1, $src2, $src3)
>;
+def : Pat <
+ (f64 (fsub f64:$src0, f64:$src1)),
+ (V_SUB_F64 $src0, $src1)
+>;
+
/********** ======================= **********/
/********** Image sampling patterns **********/
/********** ======================= **********/
def : BitConvert <f32, i32, SReg_32>;
def : BitConvert <f32, i32, VReg_32>;
+def : BitConvert <i64, f64, VReg_64>;
+
+def : BitConvert <f64, i64, VReg_64>;
+
+def : BitConvert <v2f32, v2i32, VReg_64>;
+def : BitConvert <v2i32, v2f32, VReg_64>;
+
+def : BitConvert <v4f32, v4i32, VReg_128>;
+def : BitConvert <v4i32, v4f32, VReg_128>;
+
/********** =================== **********/
/********** Src & Dst modifiers **********/
/********** =================== **********/
(S_MOV_B32 (i32 (HI32 imm:$imm))), sub1)
>;
+def : Pat <
+ (f64 fpimm:$imm),
+ (INSERT_SUBREG (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
+ (V_MOV_B32_e32 (f32 (LO32f fpimm:$imm))), sub0),
+ (V_MOV_B32_e32 (f32 (HI32f fpimm:$imm))), sub1)
+>;
+
/********** ===================== **********/
/********** Interpolation Paterns **********/
/********** ===================== **********/
(V_MUL_F32_e32 $src0, (V_RCP_F32_e32 $src1))
>;
+def : Pat<
+ (fdiv f64:$src0, f64:$src1),
+ (V_MUL_F64 $src0, (V_RCP_F64_e32 $src1), (i64 0))
+>;
+
def : Pat <
(fcos f32:$src0),
(V_COS_F32_e32 (V_MUL_F32_e32 $src0, (V_MOV_B32_e32 CONST.TWO_PI_INV)))
defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, f32>;
defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, i32>;
defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, i64>;
+defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, v2i32>;
defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, v16i8>;
defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v32i8>;
>;
}
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_ADDR64, i32,
+ sextloadi8_global, sextloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_ADDR64, i32,
+ az_extloadi8_global, az_extloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_SSHORT_ADDR64, i32,
+ sextloadi16_global, sextloadi16_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_USHORT_ADDR64, i32,
+ az_extloadi16_global, az_extloadi16_constant>;
defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORD_ADDR64, i32,
global_load, constant_load>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_ADDR64, i32,
- zextloadi8_global, zextloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, i64,
+ global_load, constant_load>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, i64,
+ az_extloadi32_global, az_extloadi32_constant>;
defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, v2i32,
global_load, constant_load>;
defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX4_ADDR64, v4i32,
// 1. Extract with offset
def : Pat<
- (vector_extract vt:$vec, (i64 (zext (add i32:$idx, imm:$off)))),
+ (vector_extract vt:$vec, (add i32:$idx, imm:$off)),
(f32 (SI_INDIRECT_SRC (IMPLICIT_DEF), $vec, $idx, imm:$off))
>;
// 2. Extract without offset
def : Pat<
- (vector_extract vt:$vec, (i64 (zext i32:$idx))),
+ (vector_extract vt:$vec, i32:$idx),
(f32 (SI_INDIRECT_SRC (IMPLICIT_DEF), $vec, $idx, 0))
>;
// 3. Insert with offset
def : Pat<
- (vector_insert vt:$vec, f32:$val, (i64 (zext (add i32:$idx, imm:$off)))),
+ (vector_insert vt:$vec, f32:$val, (add i32:$idx, imm:$off)),
(IndDst (IMPLICIT_DEF), $vec, $idx, imm:$off, $val)
>;
// 4. Insert without offset
def : Pat<
- (vector_insert vt:$vec, f32:$val, (i64 (zext i32:$idx))),
+ (vector_insert vt:$vec, f32:$val, i32:$idx),
(IndDst (IMPLICIT_DEF), $vec, $idx, 0, $val)
>;
}
projects / oota-llvm.git / blobdiff