}
def isSI : Predicate<"Subtarget.getGeneration() "
- "== AMDGPUSubtarget::SOUTHERN_ISLANDS">;
+ ">= AMDGPUSubtarget::SOUTHERN_ISLANDS">;
+
+def WAIT_FLAG : InstFlag<"printWaitFlag">;
let Predicates = [isSI] in {
def S_CMPK_LE_U32 : SOPK_32 <0x0000000e, "S_CMPK_LE_U32", []>;
} // End isCompare = 1
-def S_ADDK_I32 : SOPK_32 <0x0000000f, "S_ADDK_I32", []>;
-def S_MULK_I32 : SOPK_32 <0x00000010, "S_MULK_I32", []>;
+let Defs = [SCC], isCommutable = 1 in {
+ def S_ADDK_I32 : SOPK_32 <0x0000000f, "S_ADDK_I32", []>;
+ def S_MULK_I32 : SOPK_32 <0x00000010, "S_MULK_I32", []>;
+}
+
//def S_CBRANCH_I_FORK : SOPK_ <0x00000011, "S_CBRANCH_I_FORK", []>;
def S_GETREG_B32 : SOPK_32 <0x00000012, "S_GETREG_B32", []>;
def S_SETREG_B32 : SOPK_32 <0x00000013, "S_SETREG_B32", []>;
let isCompare = 1 in {
defm V_CMP_F_F32 : VOPC_32 <0x00000000, "V_CMP_F_F32">;
-defm V_CMP_LT_F32 : VOPC_32 <0x00000001, "V_CMP_LT_F32", f32, COND_LT>;
-defm V_CMP_EQ_F32 : VOPC_32 <0x00000002, "V_CMP_EQ_F32", f32, COND_EQ>;
-defm V_CMP_LE_F32 : VOPC_32 <0x00000003, "V_CMP_LE_F32", f32, COND_LE>;
-defm V_CMP_GT_F32 : VOPC_32 <0x00000004, "V_CMP_GT_F32", f32, COND_GT>;
-defm V_CMP_LG_F32 : VOPC_32 <0x00000005, "V_CMP_LG_F32", f32, COND_NE>;
-defm V_CMP_GE_F32 : VOPC_32 <0x00000006, "V_CMP_GE_F32", f32, COND_GE>;
-defm V_CMP_O_F32 : VOPC_32 <0x00000007, "V_CMP_O_F32">;
-defm V_CMP_U_F32 : VOPC_32 <0x00000008, "V_CMP_U_F32">;
+defm V_CMP_LT_F32 : VOPC_32 <0x00000001, "V_CMP_LT_F32", f32, COND_OLT>;
+defm V_CMP_EQ_F32 : VOPC_32 <0x00000002, "V_CMP_EQ_F32", f32, COND_OEQ>;
+defm V_CMP_LE_F32 : VOPC_32 <0x00000003, "V_CMP_LE_F32", f32, COND_OLE>;
+defm V_CMP_GT_F32 : VOPC_32 <0x00000004, "V_CMP_GT_F32", f32, COND_OGT>;
+defm V_CMP_LG_F32 : VOPC_32 <0x00000005, "V_CMP_LG_F32">;
+defm V_CMP_GE_F32 : VOPC_32 <0x00000006, "V_CMP_GE_F32", f32, COND_OGE>;
+defm V_CMP_O_F32 : VOPC_32 <0x00000007, "V_CMP_O_F32", f32, COND_O>;
+defm V_CMP_U_F32 : VOPC_32 <0x00000008, "V_CMP_U_F32", f32, COND_UO>;
defm V_CMP_NGE_F32 : VOPC_32 <0x00000009, "V_CMP_NGE_F32">;
defm V_CMP_NLG_F32 : VOPC_32 <0x0000000a, "V_CMP_NLG_F32">;
defm V_CMP_NGT_F32 : VOPC_32 <0x0000000b, "V_CMP_NGT_F32">;
defm V_CMP_NLE_F32 : VOPC_32 <0x0000000c, "V_CMP_NLE_F32">;
-defm V_CMP_NEQ_F32 : VOPC_32 <0x0000000d, "V_CMP_NEQ_F32", f32, COND_NE>;
+defm V_CMP_NEQ_F32 : VOPC_32 <0x0000000d, "V_CMP_NEQ_F32", f32, COND_UNE>;
defm V_CMP_NLT_F32 : VOPC_32 <0x0000000e, "V_CMP_NLT_F32">;
defm V_CMP_TRU_F32 : VOPC_32 <0x0000000f, "V_CMP_TRU_F32">;
} // 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", 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_LT_F64 : VOPC_64 <0x00000021, "V_CMP_LT_F64", f64, COND_OLT>;
+defm V_CMP_EQ_F64 : VOPC_64 <0x00000022, "V_CMP_EQ_F64", f64, COND_OEQ>;
+defm V_CMP_LE_F64 : VOPC_64 <0x00000023, "V_CMP_LE_F64", f64, COND_OLE>;
+defm V_CMP_GT_F64 : VOPC_64 <0x00000024, "V_CMP_GT_F64", f64, COND_OGT>;
defm V_CMP_LG_F64 : VOPC_64 <0x00000025, "V_CMP_LG_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_GE_F64 : VOPC_64 <0x00000026, "V_CMP_GE_F64", f64, COND_OGE>;
+defm V_CMP_O_F64 : VOPC_64 <0x00000027, "V_CMP_O_F64", f64, COND_O>;
+defm V_CMP_U_F64 : VOPC_64 <0x00000028, "V_CMP_U_F64", f64, COND_UO>;
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", f64, COND_NE>;
+defm V_CMP_NEQ_F64 : VOPC_64 <0x0000002d, "V_CMP_NEQ_F64", f64, COND_UNE>;
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 hasSideEffects = 1, Defs = [EXEC]
defm V_CMP_F_I32 : VOPC_32 <0x00000080, "V_CMP_F_I32">;
-defm V_CMP_LT_I32 : VOPC_32 <0x00000081, "V_CMP_LT_I32", i32, COND_LT>;
+defm V_CMP_LT_I32 : VOPC_32 <0x00000081, "V_CMP_LT_I32", i32, COND_SLT>;
defm V_CMP_EQ_I32 : VOPC_32 <0x00000082, "V_CMP_EQ_I32", i32, COND_EQ>;
-defm V_CMP_LE_I32 : VOPC_32 <0x00000083, "V_CMP_LE_I32", i32, COND_LE>;
-defm V_CMP_GT_I32 : VOPC_32 <0x00000084, "V_CMP_GT_I32", i32, COND_GT>;
+defm V_CMP_LE_I32 : VOPC_32 <0x00000083, "V_CMP_LE_I32", i32, COND_SLE>;
+defm V_CMP_GT_I32 : VOPC_32 <0x00000084, "V_CMP_GT_I32", i32, COND_SGT>;
defm V_CMP_NE_I32 : VOPC_32 <0x00000085, "V_CMP_NE_I32", i32, COND_NE>;
-defm V_CMP_GE_I32 : VOPC_32 <0x00000086, "V_CMP_GE_I32", i32, COND_GE>;
+defm V_CMP_GE_I32 : VOPC_32 <0x00000086, "V_CMP_GE_I32", i32, COND_SGE>;
defm V_CMP_T_I32 : VOPC_32 <0x00000087, "V_CMP_T_I32">;
let hasSideEffects = 1, Defs = [EXEC] in {
} // End hasSideEffects = 1, Defs = [EXEC]
defm V_CMP_F_I64 : VOPC_64 <0x000000a0, "V_CMP_F_I64">;
-defm V_CMP_LT_I64 : VOPC_64 <0x000000a1, "V_CMP_LT_I64">;
-defm V_CMP_EQ_I64 : VOPC_64 <0x000000a2, "V_CMP_EQ_I64">;
-defm V_CMP_LE_I64 : VOPC_64 <0x000000a3, "V_CMP_LE_I64">;
-defm V_CMP_GT_I64 : VOPC_64 <0x000000a4, "V_CMP_GT_I64">;
-defm V_CMP_NE_I64 : VOPC_64 <0x000000a5, "V_CMP_NE_I64">;
-defm V_CMP_GE_I64 : VOPC_64 <0x000000a6, "V_CMP_GE_I64">;
+defm V_CMP_LT_I64 : VOPC_64 <0x000000a1, "V_CMP_LT_I64", i64, COND_SLT>;
+defm V_CMP_EQ_I64 : VOPC_64 <0x000000a2, "V_CMP_EQ_I64", i64, COND_EQ>;
+defm V_CMP_LE_I64 : VOPC_64 <0x000000a3, "V_CMP_LE_I64", i64, COND_SLE>;
+defm V_CMP_GT_I64 : VOPC_64 <0x000000a4, "V_CMP_GT_I64", i64, COND_SGT>;
+defm V_CMP_NE_I64 : VOPC_64 <0x000000a5, "V_CMP_NE_I64", i64, COND_NE>;
+defm V_CMP_GE_I64 : VOPC_64 <0x000000a6, "V_CMP_GE_I64", i64, COND_SGE>;
defm V_CMP_T_I64 : VOPC_64 <0x000000a7, "V_CMP_T_I64">;
let hasSideEffects = 1, Defs = [EXEC] in {
} // End hasSideEffects = 1, Defs = [EXEC]
defm V_CMP_F_U32 : VOPC_32 <0x000000c0, "V_CMP_F_U32">;
-defm V_CMP_LT_U32 : VOPC_32 <0x000000c1, "V_CMP_LT_U32">;
-defm V_CMP_EQ_U32 : VOPC_32 <0x000000c2, "V_CMP_EQ_U32">;
-defm V_CMP_LE_U32 : VOPC_32 <0x000000c3, "V_CMP_LE_U32">;
-defm V_CMP_GT_U32 : VOPC_32 <0x000000c4, "V_CMP_GT_U32">;
-defm V_CMP_NE_U32 : VOPC_32 <0x000000c5, "V_CMP_NE_U32">;
-defm V_CMP_GE_U32 : VOPC_32 <0x000000c6, "V_CMP_GE_U32">;
+defm V_CMP_LT_U32 : VOPC_32 <0x000000c1, "V_CMP_LT_U32", i32, COND_ULT>;
+defm V_CMP_EQ_U32 : VOPC_32 <0x000000c2, "V_CMP_EQ_U32", i32, COND_EQ>;
+defm V_CMP_LE_U32 : VOPC_32 <0x000000c3, "V_CMP_LE_U32", i32, COND_ULE>;
+defm V_CMP_GT_U32 : VOPC_32 <0x000000c4, "V_CMP_GT_U32", i32, COND_UGT>;
+defm V_CMP_NE_U32 : VOPC_32 <0x000000c5, "V_CMP_NE_U32", i32, COND_NE>;
+defm V_CMP_GE_U32 : VOPC_32 <0x000000c6, "V_CMP_GE_U32", i32, COND_UGE>;
defm V_CMP_T_U32 : VOPC_32 <0x000000c7, "V_CMP_T_U32">;
let hasSideEffects = 1, Defs = [EXEC] in {
} // End hasSideEffects = 1, Defs = [EXEC]
defm V_CMP_F_U64 : VOPC_64 <0x000000e0, "V_CMP_F_U64">;
-defm V_CMP_LT_U64 : VOPC_64 <0x000000e1, "V_CMP_LT_U64">;
-defm V_CMP_EQ_U64 : VOPC_64 <0x000000e2, "V_CMP_EQ_U64">;
-defm V_CMP_LE_U64 : VOPC_64 <0x000000e3, "V_CMP_LE_U64">;
-defm V_CMP_GT_U64 : VOPC_64 <0x000000e4, "V_CMP_GT_U64">;
-defm V_CMP_NE_U64 : VOPC_64 <0x000000e5, "V_CMP_NE_U64">;
-defm V_CMP_GE_U64 : VOPC_64 <0x000000e6, "V_CMP_GE_U64">;
+defm V_CMP_LT_U64 : VOPC_64 <0x000000e1, "V_CMP_LT_U64", i64, COND_ULT>;
+defm V_CMP_EQ_U64 : VOPC_64 <0x000000e2, "V_CMP_EQ_U64", i64, COND_EQ>;
+defm V_CMP_LE_U64 : VOPC_64 <0x000000e3, "V_CMP_LE_U64", i64, COND_ULE>;
+defm V_CMP_GT_U64 : VOPC_64 <0x000000e4, "V_CMP_GT_U64", i64, COND_UGT>;
+defm V_CMP_NE_U64 : VOPC_64 <0x000000e5, "V_CMP_NE_U64", i64, COND_NE>;
+defm V_CMP_GE_U64 : VOPC_64 <0x000000e6, "V_CMP_GE_U64", i64, COND_UGE>;
defm V_CMP_T_U64 : VOPC_64 <0x000000e7, "V_CMP_T_U64">;
let hasSideEffects = 1, Defs = [EXEC] in {
} // End isCompare = 1
+def DS_ADD_U32_RTN : DS_1A1D_RET <0x20, "DS_ADD_U32_RTN", VReg_32>;
+def DS_SUB_U32_RTN : DS_1A1D_RET <0x21, "DS_SUB_U32_RTN", VReg_32>;
def DS_WRITE_B32 : DS_Store_Helper <0x0000000d, "DS_WRITE_B32", VReg_32>;
+def DS_WRITE_B8 : DS_Store_Helper <0x00000001e, "DS_WRITE_B8", VReg_32>;
+def DS_WRITE_B16 : DS_Store_Helper <0x00000001f, "DS_WRITE_B16", VReg_32>;
def DS_READ_B32 : DS_Load_Helper <0x00000036, "DS_READ_B32", VReg_32>;
+def DS_READ_I8 : DS_Load_Helper <0x00000039, "DS_READ_I8", VReg_32>;
+def DS_READ_U8 : DS_Load_Helper <0x0000003a, "DS_READ_U8", VReg_32>;
+def DS_READ_I16 : DS_Load_Helper <0x0000003b, "DS_READ_I16", VReg_32>;
+def DS_READ_U16 : DS_Load_Helper <0x0000003c, "DS_READ_U16", 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 TBUFFER_LOAD_FORMAT_XY : MTBUF_ <0x00000001, "TBUFFER_LOAD_FORMAT_XY", []>;
//def TBUFFER_LOAD_FORMAT_XYZ : MTBUF_ <0x00000002, "TBUFFER_LOAD_FORMAT_XYZ", []>;
def TBUFFER_LOAD_FORMAT_XYZW : MTBUF_Load_Helper <0x00000003, "TBUFFER_LOAD_FORMAT_XYZW", VReg_128>;
-//def TBUFFER_STORE_FORMAT_X : MTBUF_ <0x00000004, "TBUFFER_STORE_FORMAT_X", []>;
-//def TBUFFER_STORE_FORMAT_XY : MTBUF_ <0x00000005, "TBUFFER_STORE_FORMAT_XY", []>;
-//def TBUFFER_STORE_FORMAT_XYZ : MTBUF_ <0x00000006, "TBUFFER_STORE_FORMAT_XYZ", []>;
-//def TBUFFER_STORE_FORMAT_XYZW : MTBUF_ <0x00000007, "TBUFFER_STORE_FORMAT_XYZW", []>;
+def TBUFFER_STORE_FORMAT_X : MTBUF_Store_Helper <0x00000004, "TBUFFER_STORE_FORMAT_X", VReg_32>;
+def TBUFFER_STORE_FORMAT_XY : MTBUF_Store_Helper <0x00000005, "TBUFFER_STORE_FORMAT_XY", VReg_64>;
+def TBUFFER_STORE_FORMAT_XYZ : MTBUF_Store_Helper <0x00000006, "TBUFFER_STORE_FORMAT_XYZ", VReg_128>;
+def TBUFFER_STORE_FORMAT_XYZW : MTBUF_Store_Helper <0x00000007, "TBUFFER_STORE_FORMAT_XYZW", VReg_128>;
let mayLoad = 1 in {
-defm S_LOAD_DWORD : SMRD_Helper <0x00, "S_LOAD_DWORD", SReg_64, SReg_32>;
+// We are using the SGPR_32 and not the SReg_32 register class for 32-bit
+// SMRD instructions, because the SGPR_32 register class does not include M0
+// and writing to M0 from an SMRD instruction will hang the GPU.
+defm S_LOAD_DWORD : SMRD_Helper <0x00, "S_LOAD_DWORD", SReg_64, SGPR_32>;
defm S_LOAD_DWORDX2 : SMRD_Helper <0x01, "S_LOAD_DWORDX2", SReg_64, SReg_64>;
defm S_LOAD_DWORDX4 : SMRD_Helper <0x02, "S_LOAD_DWORDX4", SReg_64, SReg_128>;
defm S_LOAD_DWORDX8 : SMRD_Helper <0x03, "S_LOAD_DWORDX8", SReg_64, SReg_256>;
defm S_LOAD_DWORDX16 : SMRD_Helper <0x04, "S_LOAD_DWORDX16", SReg_64, SReg_512>;
defm S_BUFFER_LOAD_DWORD : SMRD_Helper <
- 0x08, "S_BUFFER_LOAD_DWORD", SReg_128, SReg_32
+ 0x08, "S_BUFFER_LOAD_DWORD", SReg_128, SGPR_32
>;
defm S_BUFFER_LOAD_DWORDX2 : SMRD_Helper <
//def IMAGE_STORE_MIP : MIMG_NoPattern_ <"IMAGE_STORE_MIP", 0x00000009>;
//def IMAGE_STORE_PCK : MIMG_NoPattern_ <"IMAGE_STORE_PCK", 0x0000000a>;
//def IMAGE_STORE_MIP_PCK : MIMG_NoPattern_ <"IMAGE_STORE_MIP_PCK", 0x0000000b>;
-def IMAGE_GET_RESINFO : MIMG_NoSampler_Helper <0x0000000e, "IMAGE_GET_RESINFO", VReg_32>;
+defm IMAGE_GET_RESINFO : MIMG_NoSampler <0x0000000e, "IMAGE_GET_RESINFO">;
//def IMAGE_ATOMIC_SWAP : MIMG_NoPattern_ <"IMAGE_ATOMIC_SWAP", 0x0000000f>;
//def IMAGE_ATOMIC_CMPSWAP : MIMG_NoPattern_ <"IMAGE_ATOMIC_CMPSWAP", 0x00000010>;
//def IMAGE_ATOMIC_ADD : MIMG_NoPattern_ <"IMAGE_ATOMIC_ADD", 0x00000011>;
let mayStore = 1;
}
-def S_WAITCNT : SOPP <0x0000000c, (ins i32imm:$simm16), "S_WAITCNT $simm16",
+def S_WAITCNT : SOPP <0x0000000c, (ins WAIT_FLAG:$simm16), "S_WAITCNT $simm16",
[]
>;
} // End hasSideEffects
(V_CNDMASK_B32_e64 $src0, $src1, $src2)
>;
+def : Pat <
+ (i32 (trunc i64:$val)),
+ (EXTRACT_SUBREG $val, sub0)
+>;
+
//use two V_CNDMASK_B32_e64 instructions for f64
def : Pat <
(f64 (select i1:$src2, f64:$src1, f64:$src0)),
$src2), sub1)
>;
-defm V_READLANE_B32 : VOP2_32 <0x00000001, "V_READLANE_B32", []>;
-defm V_WRITELANE_B32 : VOP2_32 <0x00000002, "V_WRITELANE_B32", []>;
+def V_READLANE_B32 : VOP2 <
+ 0x00000001,
+ (outs SReg_32:$vdst),
+ (ins VReg_32:$src0, SSrc_32:$vsrc1),
+ "V_READLANE_B32 $vdst, $src0, $vsrc1",
+ []
+>;
+
+def V_WRITELANE_B32 : VOP2 <
+ 0x00000002,
+ (outs VReg_32:$vdst),
+ (ins SReg_32:$src0, SSrc_32:$vsrc1),
+ "V_WRITELANE_B32 $vdst, $src0, $vsrc1",
+ []
+>;
let isCommutable = 1 in {
defm V_ADD_F32 : VOP2_32 <0x00000003, "V_ADD_F32",
>;
defm V_ASHRREV_I32 : VOP2_32 <0x00000018, "V_ASHRREV_I32", [], "V_ASHR_I32">;
+let hasPostISelHook = 1 in {
+
defm V_LSHL_B32 : VOP2_32 <0x00000019, "V_LSHL_B32",
[(set i32:$dst, (shl i32:$src0, i32:$src1))]
>;
+
+}
defm V_LSHLREV_B32 : VOP2_32 <0x0000001a, "V_LSHLREV_B32", [], "V_LSHL_B32">;
defm V_AND_B32 : VOP2_32 <0x0000001b, "V_AND_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",
- [(set i32:$dst, (add (i32 VSrc_32:$src0), (i32 VReg_32:$src1)))]
->;
-
-defm V_SUB_I32 : VOP2b_32 <0x00000026, "V_SUB_I32",
- [(set i32:$dst, (sub i32:$src0, i32:$src1))]
->;
+// No patterns so that the scalar instructions are always selected.
+// The scalar versions will be replaced with vector when needed later.
+defm V_ADD_I32 : VOP2b_32 <0x00000025, "V_ADD_I32", []>;
+defm V_SUB_I32 : VOP2b_32 <0x00000026, "V_SUB_I32", []>;
defm V_SUBREV_I32 : VOP2b_32 <0x00000027, "V_SUBREV_I32", [], "V_SUB_I32">;
-let Uses = [VCC] in { // Carry-out comes from VCC
+let Uses = [VCC] in { // Carry-in comes from VCC
defm V_ADDC_U32 : VOP2b_32 <0x00000028, "V_ADDC_U32", []>;
defm V_SUBB_U32 : VOP2b_32 <0x00000029, "V_SUBB_U32", []>;
defm V_SUBBREV_U32 : VOP2b_32 <0x0000002a, "V_SUBBREV_U32", [], "V_SUBB_U32">;
//def V_QSAD_U8 : VOP3_U8 <0x00000172, "V_QSAD_U8", []>;
//def V_MQSAD_U8 : VOP3_U8 <0x00000173, "V_MQSAD_U8", []>;
def V_TRIG_PREOP_F64 : VOP3_64 <0x00000174, "V_TRIG_PREOP_F64", []>;
+
+let Defs = [SCC] in { // Carry out goes to SCC
+let isCommutable = 1 in {
def S_ADD_U32 : SOP2_32 <0x00000000, "S_ADD_U32", []>;
+def S_ADD_I32 : SOP2_32 <0x00000002, "S_ADD_I32",
+ [(set i32:$dst, (add SSrc_32:$src0, SSrc_32:$src1))]
+>;
+} // End isCommutable = 1
+
def S_SUB_U32 : SOP2_32 <0x00000001, "S_SUB_U32", []>;
-def S_ADD_I32 : SOP2_32 <0x00000002, "S_ADD_I32", []>;
-def S_SUB_I32 : SOP2_32 <0x00000003, "S_SUB_I32", []>;
-def S_ADDC_U32 : SOP2_32 <0x00000004, "S_ADDC_U32", []>;
-def S_SUBB_U32 : SOP2_32 <0x00000005, "S_SUBB_U32", []>;
+def S_SUB_I32 : SOP2_32 <0x00000003, "S_SUB_I32",
+ [(set i32:$dst, (sub SSrc_32:$src0, SSrc_32:$src1))]
+>;
+
+let Uses = [SCC] in { // Carry in comes from SCC
+let isCommutable = 1 in {
+def S_ADDC_U32 : SOP2_32 <0x00000004, "S_ADDC_U32",
+ [(set i32:$dst, (adde (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>;
+} // End isCommutable = 1
+
+def S_SUBB_U32 : SOP2_32 <0x00000005, "S_SUBB_U32",
+ [(set i32:$dst, (sube (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>;
+} // End Uses = [SCC]
+} // End Defs = [SCC]
+
def S_MIN_I32 : SOP2_32 <0x00000006, "S_MIN_I32", []>;
def S_MIN_U32 : SOP2_32 <0x00000007, "S_MIN_U32", []>;
def S_MAX_I32 : SOP2_32 <0x00000008, "S_MAX_I32", []>;
def S_NOR_B64 : SOP2_64 <0x0000001b, "S_NOR_B64", []>;
def S_XNOR_B32 : SOP2_32 <0x0000001c, "S_XNOR_B32", []>;
def S_XNOR_B64 : SOP2_64 <0x0000001d, "S_XNOR_B64", []>;
-def S_LSHL_B32 : SOP2_32 <0x0000001e, "S_LSHL_B32", []>;
-def S_LSHL_B64 : SOP2_64 <0x0000001f, "S_LSHL_B64", []>;
-def S_LSHR_B32 : SOP2_32 <0x00000020, "S_LSHR_B32", []>;
-def S_LSHR_B64 : SOP2_64 <0x00000021, "S_LSHR_B64", []>;
-def S_ASHR_I32 : SOP2_32 <0x00000022, "S_ASHR_I32", []>;
-def S_ASHR_I64 : SOP2_64 <0x00000023, "S_ASHR_I64", []>;
+
+// Use added complexity so these patterns are preferred to the VALU patterns.
+let AddedComplexity = 1 in {
+
+def S_LSHL_B32 : SOP2_32 <0x0000001e, "S_LSHL_B32",
+ [(set i32:$dst, (shl i32:$src0, i32:$src1))]
+>;
+def S_LSHL_B64 : SOP2_SHIFT_64 <0x0000001f, "S_LSHL_B64",
+ [(set i64:$dst, (shl i64:$src0, i32:$src1))]
+>;
+def S_LSHR_B32 : SOP2_32 <0x00000020, "S_LSHR_B32",
+ [(set i32:$dst, (srl i32:$src0, i32:$src1))]
+>;
+def S_LSHR_B64 : SOP2_SHIFT_64 <0x00000021, "S_LSHR_B64",
+ [(set i64:$dst, (srl i64:$src0, i32:$src1))]
+>;
+def S_ASHR_I32 : SOP2_32 <0x00000022, "S_ASHR_I32",
+ [(set i32:$dst, (sra i32:$src0, i32:$src1))]
+>;
+def S_ASHR_I64 : SOP2_SHIFT_64 <0x00000023, "S_ASHR_I64",
+ [(set i64:$dst, (sra i64:$src0, i32:$src1))]
+>;
+
+} // End AddedComplexity = 1
+
def S_BFM_B32 : SOP2_32 <0x00000024, "S_BFM_B32", []>;
def S_BFM_B64 : SOP2_64 <0x00000025, "S_BFM_B64", []>;
def S_MUL_I32 : SOP2_32 <0x00000026, "S_MUL_I32", []>;
[(set GPRF32:$dst, (int_AMDGPU_load_const imm:$src))]
>;
-// SI Psuedo instructions. These are used by the CFG structurizer pass
+// SI pseudo instructions. These are used by the CFG structurizer pass
// and should be lowered to ISA instructions prior to codegen.
let mayLoad = 1, mayStore = 1, hasSideEffects = 1,
def SI_KILL : InstSI <
(outs),
(ins VReg_32:$src),
- "SI_KIL $src",
+ "SI_KILL $src",
[(int_AMDGPU_kill f32:$src)]
>;
let Uses = [EXEC], Defs = [EXEC,VCC,M0] in {
+//defm SI_ : RegisterLoadStore <VReg_32, FRAMEri64, ADDRIndirect>;
+
+let UseNamedOperandTable = 1 in {
+
+def SI_RegisterLoad : AMDGPUShaderInst <
+ (outs VReg_32:$dst, SReg_64:$temp),
+ (ins FRAMEri64:$addr, i32imm:$chan),
+ "", []
+> {
+ let isRegisterLoad = 1;
+ let mayLoad = 1;
+}
+
+class SIRegStore<dag outs> : AMDGPUShaderInst <
+ outs,
+ (ins VReg_32:$val, FRAMEri64:$addr, i32imm:$chan),
+ "", []
+> {
+ let isRegisterStore = 1;
+ let mayStore = 1;
+}
+
+let usesCustomInserter = 1 in {
+def SI_RegisterStorePseudo : SIRegStore<(outs)>;
+} // End usesCustomInserter = 1
+def SI_RegisterStore : SIRegStore<(outs SReg_64:$temp)>;
+
+
+} // End UseNamedOperandTable = 1
+
def SI_INDIRECT_SRC : InstSI <
(outs VReg_32:$dst, SReg_64:$temp),
(ins unknown:$src, VSrc_32:$idx, i32imm:$off),
let Constraints = "$src = $dst";
}
+def SI_INDIRECT_DST_V1 : SI_INDIRECT_DST<VReg_32>;
def SI_INDIRECT_DST_V2 : SI_INDIRECT_DST<VReg_64>;
def SI_INDIRECT_DST_V4 : SI_INDIRECT_DST<VReg_128>;
def SI_INDIRECT_DST_V8 : SI_INDIRECT_DST<VReg_256>;
let usesCustomInserter = 1 in {
-// This psuedo instruction takes a pointer as input and outputs a resource
+// This pseudo instruction takes a pointer as input and outputs a resource
// constant that can be used with the ADDR64 MUBUF instructions.
def SI_ADDR64_RSRC : InstSI <
(outs SReg_128:$srsrc),
/* SIsample for simple 1D texture lookup */
def : Pat <
(SIsample i32:$addr, v32i8:$rsrc, i128:$sampler, imm),
- (IMAGE_SAMPLE_V1 0xf, 0, 0, 0, 0, 0, 0, 0, $addr, $rsrc, $sampler)
+ (IMAGE_SAMPLE_V4_V1 0xf, 0, 0, 0, 0, 0, 0, 0, $addr, $rsrc, $sampler)
>;
class SamplePattern<SDNode name, MIMG opcode, ValueType vt> : Pat <
def : SampleShadowArrayPattern <SIsampled, sample_c_d, addr_type>;
}
-defm : SamplePatterns<IMAGE_SAMPLE_V2, IMAGE_SAMPLE_C_V2,
- IMAGE_SAMPLE_L_V2, IMAGE_SAMPLE_C_L_V2,
- IMAGE_SAMPLE_B_V2, IMAGE_SAMPLE_C_B_V2,
- IMAGE_SAMPLE_D_V2, IMAGE_SAMPLE_C_D_V2,
+defm : SamplePatterns<IMAGE_SAMPLE_V4_V2, IMAGE_SAMPLE_C_V4_V2,
+ IMAGE_SAMPLE_L_V4_V2, IMAGE_SAMPLE_C_L_V4_V2,
+ IMAGE_SAMPLE_B_V4_V2, IMAGE_SAMPLE_C_B_V4_V2,
+ IMAGE_SAMPLE_D_V4_V2, IMAGE_SAMPLE_C_D_V4_V2,
v2i32>;
-defm : SamplePatterns<IMAGE_SAMPLE_V4, IMAGE_SAMPLE_C_V4,
- IMAGE_SAMPLE_L_V4, IMAGE_SAMPLE_C_L_V4,
- IMAGE_SAMPLE_B_V4, IMAGE_SAMPLE_C_B_V4,
- IMAGE_SAMPLE_D_V4, IMAGE_SAMPLE_C_D_V4,
+defm : SamplePatterns<IMAGE_SAMPLE_V4_V4, IMAGE_SAMPLE_C_V4_V4,
+ IMAGE_SAMPLE_L_V4_V4, IMAGE_SAMPLE_C_L_V4_V4,
+ IMAGE_SAMPLE_B_V4_V4, IMAGE_SAMPLE_C_B_V4_V4,
+ IMAGE_SAMPLE_D_V4_V4, IMAGE_SAMPLE_C_D_V4_V4,
v4i32>;
-defm : SamplePatterns<IMAGE_SAMPLE_V8, IMAGE_SAMPLE_C_V8,
- IMAGE_SAMPLE_L_V8, IMAGE_SAMPLE_C_L_V8,
- IMAGE_SAMPLE_B_V8, IMAGE_SAMPLE_C_B_V8,
- IMAGE_SAMPLE_D_V8, IMAGE_SAMPLE_C_D_V8,
+defm : SamplePatterns<IMAGE_SAMPLE_V4_V8, IMAGE_SAMPLE_C_V4_V8,
+ IMAGE_SAMPLE_L_V4_V8, IMAGE_SAMPLE_C_L_V4_V8,
+ IMAGE_SAMPLE_B_V4_V8, IMAGE_SAMPLE_C_B_V4_V8,
+ IMAGE_SAMPLE_D_V4_V8, IMAGE_SAMPLE_C_D_V4_V8,
v8i32>;
-defm : SamplePatterns<IMAGE_SAMPLE_V16, IMAGE_SAMPLE_C_V16,
- IMAGE_SAMPLE_L_V16, IMAGE_SAMPLE_C_L_V16,
- IMAGE_SAMPLE_B_V16, IMAGE_SAMPLE_C_B_V16,
- IMAGE_SAMPLE_D_V16, IMAGE_SAMPLE_C_D_V16,
+defm : SamplePatterns<IMAGE_SAMPLE_V4_V16, IMAGE_SAMPLE_C_V4_V16,
+ IMAGE_SAMPLE_L_V4_V16, IMAGE_SAMPLE_C_L_V4_V16,
+ IMAGE_SAMPLE_B_V4_V16, IMAGE_SAMPLE_C_B_V4_V16,
+ IMAGE_SAMPLE_D_V4_V16, IMAGE_SAMPLE_C_D_V4_V16,
v16i32>;
/* int_SI_imageload for texture fetches consuming varying address parameters */
def : ImageLoadArrayMSAAPattern <int_SI_imageload, opcode, addr_type>;
}
-defm : ImageLoadPatterns<IMAGE_LOAD_MIP_V2, v2i32>;
-defm : ImageLoadPatterns<IMAGE_LOAD_MIP_V4, v4i32>;
+defm : ImageLoadPatterns<IMAGE_LOAD_MIP_V4_V2, v2i32>;
+defm : ImageLoadPatterns<IMAGE_LOAD_MIP_V4_V4, v4i32>;
-defm : ImageLoadMSAAPatterns<IMAGE_LOAD_V2, v2i32>;
-defm : ImageLoadMSAAPatterns<IMAGE_LOAD_V4, v4i32>;
+defm : ImageLoadMSAAPatterns<IMAGE_LOAD_V4_V2, v2i32>;
+defm : ImageLoadMSAAPatterns<IMAGE_LOAD_V4_V4, v4i32>;
/* Image resource information */
def : Pat <
(int_SI_resinfo i32:$mipid, v32i8:$rsrc, imm),
- (IMAGE_GET_RESINFO 0xf, 0, 0, 0, 0, 0, 0, 0, (V_MOV_B32_e32 $mipid), $rsrc)
+ (IMAGE_GET_RESINFO_V4_V1 0xf, 0, 0, 0, 0, 0, 0, 0, (V_MOV_B32_e32 $mipid), $rsrc)
>;
def : Pat <
(int_SI_resinfo i32:$mipid, v32i8:$rsrc, TEX_ARRAY),
- (IMAGE_GET_RESINFO 0xf, 0, 0, 1, 0, 0, 0, 0, (V_MOV_B32_e32 $mipid), $rsrc)
+ (IMAGE_GET_RESINFO_V4_V1 0xf, 0, 0, 1, 0, 0, 0, 0, (V_MOV_B32_e32 $mipid), $rsrc)
>;
def : Pat <
(int_SI_resinfo i32:$mipid, v32i8:$rsrc, TEX_ARRAY_MSAA),
- (IMAGE_GET_RESINFO 0xf, 0, 0, 1, 0, 0, 0, 0, (V_MOV_B32_e32 $mipid), $rsrc)
+ (IMAGE_GET_RESINFO_V4_V1 0xf, 0, 0, 1, 0, 0, 0, 0, (V_MOV_B32_e32 $mipid), $rsrc)
>;
/********** ============================================ **********/
def : BitConvert <v2f32, v2i32, VReg_64>;
def : BitConvert <v2i32, v2f32, VReg_64>;
+def : BitConvert <v2i32, i64, VReg_64>;
def : BitConvert <v4f32, v4i32, VReg_128>;
def : BitConvert <v4i32, v4f32, VReg_128>;
+def : BitConvert <v4i32, i128, VReg_128>;
+def : BitConvert <i128, v4i32, VReg_128>;
def : BitConvert <v8i32, v32i8, SReg_256>;
def : BitConvert <v32i8, v8i32, SReg_256>;
/********** Load/Store Patterns **********/
/********** ======================= **********/
-def : Pat <
- (local_load i64:$src0),
- (i32 (DS_READ_B32 0, (EXTRACT_SUBREG $src0, sub0),
- (EXTRACT_SUBREG $src0, sub0), (EXTRACT_SUBREG $src0, sub0), 0, 0))
+class DSReadPat <DS inst, ValueType vt, PatFrag frag> : Pat <
+ (frag i32:$src0),
+ (vt (inst 0, $src0, $src0, $src0, 0, 0))
>;
+def : DSReadPat <DS_READ_I8, i32, sextloadi8_local>;
+def : DSReadPat <DS_READ_U8, i32, az_extloadi8_local>;
+def : DSReadPat <DS_READ_I16, i32, sextloadi16_local>;
+def : DSReadPat <DS_READ_U16, i32, az_extloadi16_local>;
+def : DSReadPat <DS_READ_B32, i32, local_load>;
def : Pat <
- (local_store i32:$src1, i64:$src0),
- (DS_WRITE_B32 0, (EXTRACT_SUBREG $src0, sub0), $src1, $src1, 0, 0)
+ (local_load i32:$src0),
+ (i32 (DS_READ_B32 0, $src0, $src0, $src0, 0, 0))
>;
+class DSWritePat <DS inst, ValueType vt, PatFrag frag> : Pat <
+ (frag i32:$src1, i32:$src0),
+ (inst 0, $src0, $src1, $src1, 0, 0)
+>;
+
+def : DSWritePat <DS_WRITE_B8, i32, truncstorei8_local>;
+def : DSWritePat <DS_WRITE_B16, i32, truncstorei16_local>;
+def : DSWritePat <DS_WRITE_B32, i32, local_store>;
+
+def : Pat <(atomic_load_add_local i32:$ptr, i32:$val),
+ (DS_ADD_U32_RTN 0, $ptr, $val, 0, 0)>;
+
+def : Pat <(atomic_load_sub_local i32:$ptr, i32:$val),
+ (DS_SUB_U32_RTN 0, $ptr, $val, 0, 0)>;
+
/********** ================== **********/
/********** SMRD Patterns **********/
/********** ================== **********/
defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, i64>;
defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, v2i32>;
defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, i128>;
+defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, v4i32>;
defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v32i8>;
+defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v8i32>;
+defm : SMRD_Pattern <S_LOAD_DWORDX16_IMM, S_LOAD_DWORDX16_SGPR, v16i32>;
//===----------------------------------------------------------------------===//
// MUBUF Patterns
defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, v2i32, global_store>;
defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX4, v4i32, global_store>;
+//===----------------------------------------------------------------------===//
+// MTBUF Patterns
+//===----------------------------------------------------------------------===//
+
+// TBUFFER_STORE_FORMAT_*, addr64=0
+class MTBUF_StoreResource <ValueType vt, int num_channels, MTBUF opcode> : Pat<
+ (SItbuffer_store i128:$rsrc, vt:$vdata, num_channels, i32:$vaddr,
+ i32:$soffset, imm:$inst_offset, imm:$dfmt,
+ imm:$nfmt, imm:$offen, imm:$idxen,
+ imm:$glc, imm:$slc, imm:$tfe),
+ (opcode
+ $vdata, (as_i16imm $inst_offset), (as_i1imm $offen), (as_i1imm $idxen),
+ (as_i1imm $glc), 0, (as_i8imm $dfmt), (as_i8imm $nfmt), $vaddr, $rsrc,
+ (as_i1imm $slc), (as_i1imm $tfe), $soffset)
+>;
+
+def : MTBUF_StoreResource <i32, 1, TBUFFER_STORE_FORMAT_X>;
+def : MTBUF_StoreResource <v2i32, 2, TBUFFER_STORE_FORMAT_XY>;
+def : MTBUF_StoreResource <v4i32, 3, TBUFFER_STORE_FORMAT_XYZ>;
+def : MTBUF_StoreResource <v4i32, 4, TBUFFER_STORE_FORMAT_XYZW>;
+
/********** ====================== **********/
/********** Indirect adressing **********/
/********** ====================== **********/
(V_CMP_U_F32_e64 $src0, $src1)
>;
+//===----------------------------------------------------------------------===//
+// Miscellaneous Patterns
+//===----------------------------------------------------------------------===//
+
+def : Pat <
+ (i64 (trunc i128:$x)),
+ (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+ (i32 (EXTRACT_SUBREG $x, sub0)), sub0),
+ (i32 (EXTRACT_SUBREG $x, sub1)), sub1)
+>;
+
+def : Pat <
+ (i32 (trunc i64:$a)),
+ (EXTRACT_SUBREG $a, sub0)
+>;
+
+// V_ADD_I32_e32/S_ADD_I32 produces carry in VCC/SCC. For the vector
+// case, the sgpr-copies pass will fix this to use the vector version.
+def : Pat <
+ (i32 (addc i32:$src0, i32:$src1)),
+ (S_ADD_I32 $src0, $src1)
+>;
+
+def : Pat <
+ (or i64:$a, i64:$b),
+ (INSERT_SUBREG
+ (INSERT_SUBREG (IMPLICIT_DEF),
+ (V_OR_B32_e32 (EXTRACT_SUBREG $a, sub0), (EXTRACT_SUBREG $b, sub0)), sub0),
+ (V_OR_B32_e32 (EXTRACT_SUBREG $a, sub1), (EXTRACT_SUBREG $b, sub1)), sub1)
+>;
+
//============================================================================//
// Miscellaneous Optimization Patterns
//============================================================================//
projects / oota-llvm.git / blobdiff