let PrintMethod = "printInterpSlot";
}
-def isSI : Predicate<"Subtarget.device()"
- "->getGeneration() == AMDGPUDeviceInfo::HD7XXX">;
+def isSI : Predicate<"Subtarget.getGeneration() "
+ "== AMDGPUSubtarget::SOUTHERN_ISLANDS">;
let Predicates = [isSI] in {
} // 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">;
} // End isCompare = 1
+def DS_WRITE_B32 : DS_Store_Helper <0x0000000d, "DS_WRITE_B32", VReg_32>;
+def DS_READ_B32 : DS_Load_Helper <0x00000036, "DS_READ_B32", VReg_32>;
+
//def BUFFER_LOAD_FORMAT_X : MUBUF_ <0x00000000, "BUFFER_LOAD_FORMAT_X", []>;
//def BUFFER_LOAD_FORMAT_XY : MUBUF_ <0x00000001, "BUFFER_LOAD_FORMAT_XY", []>;
//def BUFFER_LOAD_FORMAT_XYZ : MUBUF_ <0x00000002, "BUFFER_LOAD_FORMAT_XYZ", []>;
-def BUFFER_LOAD_FORMAT_XYZW : MUBUF_Load_Helper <0x00000003, "BUFFER_LOAD_FORMAT_XYZW", VReg_128>;
+defm BUFFER_LOAD_FORMAT_XYZW : MUBUF_Load_Helper <0x00000003, "BUFFER_LOAD_FORMAT_XYZW", VReg_128>;
//def BUFFER_STORE_FORMAT_X : MUBUF_ <0x00000004, "BUFFER_STORE_FORMAT_X", []>;
//def BUFFER_STORE_FORMAT_XY : MUBUF_ <0x00000005, "BUFFER_STORE_FORMAT_XY", []>;
//def BUFFER_STORE_FORMAT_XYZ : MUBUF_ <0x00000006, "BUFFER_STORE_FORMAT_XYZ", []>;
//def BUFFER_STORE_FORMAT_XYZW : MUBUF_ <0x00000007, "BUFFER_STORE_FORMAT_XYZW", []>;
-//def BUFFER_LOAD_UBYTE : MUBUF_ <0x00000008, "BUFFER_LOAD_UBYTE", []>;
-//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", []>;
-def BUFFER_LOAD_DWORD : MUBUF_Load_Helper <0x0000000c, "BUFFER_LOAD_DWORD", VReg_32>;
-def BUFFER_LOAD_DWORDX2 : MUBUF_Load_Helper <0x0000000d, "BUFFER_LOAD_DWORDX2", VReg_64>;
-def BUFFER_LOAD_DWORDX4 : MUBUF_Load_Helper <0x0000000e, "BUFFER_LOAD_DWORDX4", VReg_128>;
+defm BUFFER_LOAD_UBYTE : MUBUF_Load_Helper <0x00000008, "BUFFER_LOAD_UBYTE", VReg_32>;
+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>;
//def BUFFER_STORE_BYTE : MUBUF_ <0x00000018, "BUFFER_STORE_BYTE", []>;
//def BUFFER_STORE_SHORT : MUBUF_ <0x0000001a, "BUFFER_STORE_SHORT", []>;
def BUFFER_STORE_DWORDX2 : MUBUF_Store_Helper <
0x0000001d, "BUFFER_STORE_DWORDX2", VReg_64, i64
>;
-//def BUFFER_STORE_DWORDX4 : MUBUF_DWORDX4 <0x0000001e, "BUFFER_STORE_DWORDX4", []>;
+
+def BUFFER_STORE_DWORDX4 : MUBUF_Store_Helper <
+ 0x0000001e, "BUFFER_STORE_DWORDX4", VReg_128, v4i32
+>;
//def BUFFER_ATOMIC_SWAP : MUBUF_ <0x00000030, "BUFFER_ATOMIC_SWAP", []>;
//def BUFFER_ATOMIC_CMPSWAP : MUBUF_ <0x00000031, "BUFFER_ATOMIC_CMPSWAP", []>;
//def BUFFER_ATOMIC_ADD : MUBUF_ <0x00000032, "BUFFER_ATOMIC_ADD", []>;
//def IMAGE_SAMPLE_LZ : MIMG_NoPattern_ <"IMAGE_SAMPLE_LZ", 0x00000027>;
def IMAGE_SAMPLE_C : MIMG_Sampler_Helper <0x00000028, "IMAGE_SAMPLE_C">;
//def IMAGE_SAMPLE_C_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CL", 0x00000029>;
-//def IMAGE_SAMPLE_C_D : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D", 0x0000002a>;
+def IMAGE_SAMPLE_C_D : MIMG_Sampler_Helper <0x0000002a, "IMAGE_SAMPLE_C_D">;
//def IMAGE_SAMPLE_C_D_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_CL", 0x0000002b>;
def IMAGE_SAMPLE_C_L : MIMG_Sampler_Helper <0x0000002c, "IMAGE_SAMPLE_C_L">;
def IMAGE_SAMPLE_C_B : MIMG_Sampler_Helper <0x0000002d, "IMAGE_SAMPLE_C_B">;
} // 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", []>;
} // End isBranch = 1
} // End isTerminator = 1
-//def S_BARRIER : SOPP_ <0x0000000a, "S_BARRIER", []>;
let hasSideEffects = 1 in {
+def S_BARRIER : SOPP <0x0000000a, (ins), "S_BARRIER",
+ [(int_AMDGPU_barrier_local)]
+> {
+ let SIMM16 = 0;
+ let isBarrier = 1;
+ let hasCtrlDep = 1;
+ let mayLoad = 1;
+ let mayStore = 1;
+}
+
def S_WAITCNT : SOPP <0x0000000c, (ins i32imm:$simm16), "S_WAITCNT $simm16",
[]
>;
(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:$src1))]
defm V_MADMK_F32 : VOP2_32 <0x00000020, "V_MADMK_F32", []>;
defm V_MADAK_F32 : VOP2_32 <0x00000021, "V_MADAK_F32", []>;
//defm V_BCNT_U32_B32 : VOP2_32 <0x00000022, "V_BCNT_U32_B32", []>;
-//defm V_MBCNT_LO_U32_B32 : VOP2_32 <0x00000023, "V_MBCNT_LO_U32_B32", []>;
-//defm V_MBCNT_HI_U32_B32 : VOP2_32 <0x00000024, "V_MBCNT_HI_U32_B32", []>;
+defm V_MBCNT_LO_U32_B32 : VOP2_32 <0x00000023, "V_MBCNT_LO_U32_B32", []>;
+defm V_MBCNT_HI_U32_B32 : VOP2_32 <0x00000024, "V_MBCNT_HI_U32_B32", []>;
let isCommutable = 1, Defs = [VCC] in { // Carry-out goes to VCC
defm V_ADD_I32 : VOP2b_32 <0x00000025, "V_ADD_I32",
def V_MAD_LEGACY_F32 : VOP3_32 <0x00000140, "V_MAD_LEGACY_F32", []>;
def V_MAD_F32 : VOP3_32 <0x00000141, "V_MAD_F32", []>;
-//def V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24", []>;
-//def V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24", []>;
+def V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24",
+ [(set i32:$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:$src2))]
+>;
} // End neverHasSideEffects
def V_CUBEID_F32 : VOP3_32 <0x00000144, "V_CUBEID_F32", []>;
def V_FMA_F64 : VOP3_64 <0x0000014c, "V_FMA_F64", []>;
//def V_LERP_U8 : VOP3_U8 <0x0000014d, "V_LERP_U8", []>;
def V_ALIGNBIT_B32 : VOP3_32 <0x0000014e, "V_ALIGNBIT_B32", []>;
+def : ROTRPattern <V_ALIGNBIT_B32>;
+
def V_ALIGNBYTE_B32 : VOP3_32 <0x0000014f, "V_ALIGNBYTE_B32", []>;
def V_MULLIT_F32 : VOP3_32 <0x00000150, "V_MULLIT_F32", []>;
////def V_MIN3_F32 : VOP3_MIN3 <0x00000151, "V_MIN3_F32", []>;
def V_LSHR_B64 : VOP3_64_Shift <0x00000162, "V_LSHR_B64",
[(set i64:$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, $src1, (i64 0))
+>;
+
def V_LDEXP_F64 : VOP3_64 <0x00000168, "V_LDEXP_F64", []>;
let isCommutable = 1 in {
} // Uses = [EXEC,VCC,M0], Defs = [EXEC,VCC,M0]
+let usesCustomInserter = 1 in {
+
+// This psuedo instruction takes a pointer as input and outputs a resource
+// constant that can be used with the ADDR64 MUBUF instructions.
+def SI_ADDR64_RSRC : InstSI <
+ (outs SReg_128:$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, $src1",
+ []
+>;
+
+} // end usesCustomInserter
+
} // end IsCodeGenOnly, isPseudo
def : Pat<
/* int_SI_vs_load_input */
def : Pat<
- (int_SI_vs_load_input v16i8:$tlst, IMM12bit:$attr_offset,
- i32:$buf_idx_vgpr),
- (BUFFER_LOAD_FORMAT_XYZW imm:$attr_offset, 0, 1, 0, 0, 0,
- $buf_idx_vgpr, $tlst, 0, 0, 0)
+ (int_SI_vs_load_input v16i8:$tlst, IMM12bit:$attr_offset, i32:$buf_idx_vgpr),
+ (BUFFER_LOAD_FORMAT_XYZW_IDXEN $tlst, $buf_idx_vgpr, imm:$attr_offset)
>;
/* int_SI_export */
$src0, $src1, $src2, $src3)
>;
+def : Pat <
+ (f64 (fsub f64:$src0, f64:$src1)),
+ (V_SUB_F64 $src0, $src1)
+>;
+
/********** ======================= **********/
/********** Image sampling patterns **********/
/********** ======================= **********/
def : SampleArrayPattern <int_SI_sampleb, IMAGE_SAMPLE_B, addr_type>;
def : SampleShadowPattern <int_SI_sampleb, IMAGE_SAMPLE_C_B, addr_type>;
def : SampleShadowArrayPattern <int_SI_sampleb, IMAGE_SAMPLE_C_B, addr_type>;
+
+ def : SamplePattern <int_SI_sampled, IMAGE_SAMPLE_D, addr_type>;
+ def : SampleArrayPattern <int_SI_sampled, IMAGE_SAMPLE_D, addr_type>;
+ def : SampleShadowPattern <int_SI_sampled, IMAGE_SAMPLE_C_D, addr_type>;
+ def : SampleShadowArrayPattern <int_SI_sampled, IMAGE_SAMPLE_C_D, addr_type>;
}
defm : SamplePatterns<v2i32>;
def : BitConvert <f32, i32, SReg_32>;
def : BitConvert <f32, i32, VReg_32>;
+def : BitConvert <i64, f64, VReg_64>;
+
+def : BitConvert <f64, i64, VReg_64>;
+
+def : BitConvert <v2f32, v2i32, VReg_64>;
+def : BitConvert <v2i32, v2f32, VReg_64>;
+
+def : BitConvert <v4f32, v4i32, VReg_128>;
+def : BitConvert <v4i32, v4f32, VReg_128>;
+
/********** =================== **********/
/********** Src & Dst modifiers **********/
/********** =================== **********/
(S_MOV_B32 (i32 (HI32 imm:$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 $src0, (V_MOV_B32_e32 CONST.TWO_PI_INV)))
// 3. Offset in an 32Bit VGPR
def : Pat <
(int_SI_load_const v16i8:$sbase, i32:$voff),
- (BUFFER_LOAD_DWORD 0, 1, 0, 0, 0, 0, $voff, $sbase, 0, 0, 0)
+ (BUFFER_LOAD_DWORD_OFFEN $sbase, $voff)
>;
// The multiplication scales from [0,1] to the unsigned integer range
(V_RCP_IFLAG_F32_e32 (V_CVT_F32_U32_e32 $src0))))
>;
+def : Pat <
+ (int_SI_tid),
+ (V_MBCNT_HI_U32_B32_e32 0xffffffff,
+ (V_MBCNT_LO_U32_B32_e64 0xffffffff, 0, 0, 0, 0, 0))
+>;
+
/********** ================== **********/
/********** VOP3 Patterns **********/
/********** ================== **********/
(V_MAD_F32 $src0, $src1, $src2)
>;
+/********** ======================= **********/
+/********** Load/Store Patterns **********/
+/********** ======================= **********/
+
+def : Pat <
+ (local_load i64:$src0),
+ (i32 (DS_READ_B32 0, (EXTRACT_SUBREG $src0, sub0),
+ (EXTRACT_SUBREG $src0, sub0), (EXTRACT_SUBREG $src0, sub0), 0, 0))
+>;
+
+def : Pat <
+ (local_store i32:$src1, i64:$src0),
+ (DS_WRITE_B32 0, (EXTRACT_SUBREG $src0, sub0), $src1, $src1, 0, 0)
+>;
+
/********** ================== **********/
/********** SMRD Patterns **********/
/********** ================== **********/
defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, f32>;
defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, i32>;
+defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, i64>;
+defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, v2i32>;
defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, v16i8>;
defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v32i8>;
+//===----------------------------------------------------------------------===//
+// MUBUF Patterns
+//===----------------------------------------------------------------------===//
+
+multiclass MUBUFLoad_Pattern <MUBUF Instr_ADDR64, ValueType vt,
+ PatFrag global_ld, PatFrag constant_ld> {
+ def : Pat <
+ (vt (global_ld (add i64:$ptr, (i64 IMM12bit:$offset)))),
+ (Instr_ADDR64 (SI_ADDR64_RSRC (i64 0)), $ptr, (as_i16imm $offset))
+ >;
+
+ def : Pat <
+ (vt (global_ld i64:$ptr)),
+ (Instr_ADDR64 (SI_ADDR64_RSRC (i64 0)), $ptr, 0)
+ >;
+
+ def : Pat <
+ (vt (global_ld (add i64:$ptr, i64:$offset))),
+ (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, 0)
+ >;
+
+ def : Pat <
+ (vt (constant_ld (add i64:$ptr, i64:$offset))),
+ (Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, 0)
+ >;
+}
+
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_ADDR64, i32,
+ sextloadi8_global, sextloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_ADDR64, i32,
+ az_extloadi8_global, az_extloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_SSHORT_ADDR64, i32,
+ sextloadi16_global, sextloadi16_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_USHORT_ADDR64, i32,
+ az_extloadi16_global, az_extloadi16_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORD_ADDR64, i32,
+ global_load, constant_load>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, i64,
+ global_load, constant_load>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, i64,
+ az_extloadi32_global, az_extloadi32_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, v2i32,
+ global_load, constant_load>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX4_ADDR64, v4i32,
+ global_load, constant_load>;
+
+multiclass MUBUFStore_Pattern <MUBUF Instr, ValueType vt> {
+
+ def : Pat <
+ (global_store vt:$value, i64:$ptr),
+ (Instr $value, (SI_ADDR64_RSRC (i64 0)), $ptr, 0)
+ >;
+
+ def : Pat <
+ (global_store vt:$value, (add i64:$ptr, i64:$offset)),
+ (Instr $value, (SI_ADDR64_RSRC $ptr), $offset, 0)
+ >;
+}
+
+defm : MUBUFStore_Pattern <BUFFER_STORE_DWORD, i32>;
+defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, i64>;
+defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, v2i32>;
+defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX4, v4i32>;
+
/********** ====================== **********/
/********** Indirect adressing **********/
/********** ====================== **********/
// 1. Extract with offset
def : Pat<
- (vector_extract vt:$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)
>;
}