Hexagon: Paper over the undefined behaviour introduced by r238692

[oota-llvm.git] / lib / Target / Hexagon / HexagonIntrinsicsV4.td

diff --git a/lib/Target/Hexagon/HexagonIntrinsicsV4.td b/lib/Target/Hexagon/HexagonIntrinsicsV4.td
index ae2d463fa68a693fced7c112ff68f2849c3ebb06..c80a188d82e7ed93cbfb03fe425d901761313997 100644 (file)
--- a/lib/Target/Hexagon/HexagonIntrinsicsV4.td
+++ b/lib/Target/Hexagon/HexagonIntrinsicsV4.td
@@ -12,6 +12,38 @@
 // 80-V9418-12 Rev. A
 // June 15, 2010
 
+// Vector reduce multiply word by signed half (32x16)
+//Rdd=vrmpyweh(Rss,Rtt)[:<<1]
+def : T_PP_pat <M4_vrmpyeh_s0, int_hexagon_M4_vrmpyeh_s0>;
+def : T_PP_pat <M4_vrmpyeh_s1, int_hexagon_M4_vrmpyeh_s1>;
+
+//Rdd=vrmpywoh(Rss,Rtt)[:<<1]
+def : T_PP_pat <M4_vrmpyoh_s0, int_hexagon_M4_vrmpyoh_s0>;
+def : T_PP_pat <M4_vrmpyoh_s1, int_hexagon_M4_vrmpyoh_s1>;
+
+//Rdd+=vrmpyweh(Rss,Rtt)[:<<1]
+def : T_PPP_pat <M4_vrmpyeh_acc_s0, int_hexagon_M4_vrmpyeh_acc_s0>;
+def : T_PPP_pat <M4_vrmpyeh_acc_s1, int_hexagon_M4_vrmpyeh_acc_s1>;
+
+//Rdd=vrmpywoh(Rss,Rtt)[:<<1]
+def : T_PPP_pat <M4_vrmpyoh_acc_s0, int_hexagon_M4_vrmpyoh_acc_s0>;
+def : T_PPP_pat <M4_vrmpyoh_acc_s1, int_hexagon_M4_vrmpyoh_acc_s1>;
+
+// Vector multiply halfwords, signed by unsigned
+// Rdd=vmpyhsu(Rs,Rt)[:<<1]:sat
+def : T_RR_pat <M2_vmpy2su_s0, int_hexagon_M2_vmpy2su_s0>;
+def : T_RR_pat <M2_vmpy2su_s1, int_hexagon_M2_vmpy2su_s1>;
+
+// Rxx+=vmpyhsu(Rs,Rt)[:<<1]:sat
+def : T_PRR_pat <M2_vmac2su_s0, int_hexagon_M2_vmac2su_s0>;
+def : T_PRR_pat <M2_vmac2su_s1, int_hexagon_M2_vmac2su_s1>;
+
+// Vector polynomial multiply halfwords
+// Rdd=vpmpyh(Rs,Rt)
+def : T_RR_pat <M4_vpmpyh, int_hexagon_M4_vpmpyh>;
+// Rxx[^]=vpmpyh(Rs,Rt)
+def : T_PRR_pat <M4_vpmpyh_acc, int_hexagon_M4_vpmpyh_acc>;
+
 // Polynomial multiply words
 // Rdd=pmpyw(Rs,Rt)
 def : T_RR_pat <M4_pmpyw, int_hexagon_M4_pmpyw>;
@@ -46,6 +78,25 @@ def: T_RR_pat<C4_nbitsset,  int_hexagon_C4_nbitsset>;
 def: T_RR_pat<C4_nbitsclr,  int_hexagon_C4_nbitsclr>;
 def: T_RI_pat<C4_nbitsclri, int_hexagon_C4_nbitsclri>;
 
+
+class vcmpImm_pat <InstHexagon MI, Intrinsic IntID, PatLeaf immPred> :
+      Pat <(IntID  (i64 DoubleRegs:$src1), immPred:$src2),
+           (MI (i64 DoubleRegs:$src1), immPred:$src2)>;
+
+def : vcmpImm_pat <A4_vcmpbeqi, int_hexagon_A4_vcmpbeqi, u8ImmPred>;
+def : vcmpImm_pat <A4_vcmpbgti, int_hexagon_A4_vcmpbgti, s8ImmPred>;
+def : vcmpImm_pat <A4_vcmpbgtui, int_hexagon_A4_vcmpbgtui, u7ImmPred>;
+
+def : vcmpImm_pat <A4_vcmpheqi, int_hexagon_A4_vcmpheqi, s8ImmPred>;
+def : vcmpImm_pat <A4_vcmphgti, int_hexagon_A4_vcmphgti, s8ImmPred>;
+def : vcmpImm_pat <A4_vcmphgtui, int_hexagon_A4_vcmphgtui, u7ImmPred>;
+
+def : vcmpImm_pat <A4_vcmpweqi, int_hexagon_A4_vcmpweqi, s8ImmPred>;
+def : vcmpImm_pat <A4_vcmpwgti, int_hexagon_A4_vcmpwgti, s8ImmPred>;
+def : vcmpImm_pat <A4_vcmpwgtui, int_hexagon_A4_vcmpwgtui, u7ImmPred>;
+
+def : T_PP_pat<A4_vcmpbeq_any, int_hexagon_A4_vcmpbeq_any>;
+
 def : T_RR_pat<A4_cmpbeq,   int_hexagon_A4_cmpbeq>;
 def : T_RR_pat<A4_cmpbgt,   int_hexagon_A4_cmpbgt>;
 def : T_RR_pat<A4_cmpbgtu,  int_hexagon_A4_cmpbgtu>;
@@ -77,9 +128,25 @@ def : T_RRI_pat <M4_mpyri_addr, int_hexagon_M4_mpyri_addr>;
 def : T_RRR_pat <M4_mac_up_s1_sat, int_hexagon_M4_mac_up_s1_sat>;
 def : T_RRR_pat <M4_nac_up_s1_sat, int_hexagon_M4_nac_up_s1_sat>;
 
+// Complex multiply 32x16
+def : T_PR_pat <M4_cmpyi_wh, int_hexagon_M4_cmpyi_wh>;
+def : T_PR_pat <M4_cmpyr_wh, int_hexagon_M4_cmpyr_wh>;
+
+def : T_PR_pat <M4_cmpyi_whc, int_hexagon_M4_cmpyi_whc>;
+def : T_PR_pat <M4_cmpyr_whc, int_hexagon_M4_cmpyr_whc>;
+
 def : T_PP_pat<A4_andnp, int_hexagon_A4_andnp>;
 def : T_PP_pat<A4_ornp,  int_hexagon_A4_ornp>;
 
+// Complex add/sub halfwords/words
+def : T_PP_pat <S4_vxaddsubw, int_hexagon_S4_vxaddsubw>;
+def : T_PP_pat <S4_vxsubaddw, int_hexagon_S4_vxsubaddw>;
+def : T_PP_pat <S4_vxaddsubh, int_hexagon_S4_vxaddsubh>;
+def : T_PP_pat <S4_vxsubaddh, int_hexagon_S4_vxsubaddh>;
+
+def : T_PP_pat <S4_vxaddsubhr, int_hexagon_S4_vxaddsubhr>;
+def : T_PP_pat <S4_vxsubaddhr, int_hexagon_S4_vxsubaddhr>;
+
 // Extract bitfield
 def : T_PP_pat  <S4_extractp_rp, int_hexagon_S4_extractp_rp>;
 def : T_RP_pat  <S4_extract_rp, int_hexagon_S4_extract_rp>;
@@ -109,6 +176,18 @@ def : T_PPR_pat <A4_vrminuh, int_hexagon_A4_vrminuh>;
 def : T_PPR_pat <A4_vrminw, int_hexagon_A4_vrminw>;
 def : T_PPR_pat <A4_vrminuw, int_hexagon_A4_vrminuw>;
 
+// Rotate and reduce bytes
+def : Pat <(int_hexagon_S4_vrcrotate DoubleRegs:$src1, IntRegs:$src2,
+                                     u2ImmPred:$src3),
+           (S4_vrcrotate DoubleRegs:$src1, IntRegs:$src2, u2ImmPred:$src3)>;
+
+// Rotate and reduce bytes with accumulation
+// Rxx+=vrcrotate(Rss,Rt,#u2)
+def : Pat <(int_hexagon_S4_vrcrotate_acc DoubleRegs:$src1, DoubleRegs:$src2,
+                                         IntRegs:$src3, u2ImmPred:$src4),
+           (S4_vrcrotate_acc DoubleRegs:$src1, DoubleRegs:$src2,
+                             IntRegs:$src3, u2ImmPred:$src4)>;
+
 // Vector conditional negate
 def : T_PPR_pat<S2_vrcnegh, int_hexagon_S2_vrcnegh>;
 
@@ -155,17 +234,17 @@ def: T_RR_pat<A4_orn,  int_hexagon_A4_orn>;
 *********************************************************************/
 
 // Combine Words Into Doublewords.
-def: T_RI_pat<A4_combineri, int_hexagon_A4_combineri, s8ExtPred>;
-def: T_IR_pat<A4_combineir, int_hexagon_A4_combineir, s8ExtPred>;
+def: T_RI_pat<A4_combineri, int_hexagon_A4_combineri, s32ImmPred>;
+def: T_IR_pat<A4_combineir, int_hexagon_A4_combineir, s32ImmPred>;
 
 /********************************************************************
 *            ALU32/PRED                                             *
 *********************************************************************/
 
 // Compare
-def : T_RI_pat<C4_cmpneqi, int_hexagon_C4_cmpneqi, s10ExtPred>;
-def : T_RI_pat<C4_cmpltei, int_hexagon_C4_cmpltei, s10ExtPred>;
-def : T_RI_pat<C4_cmplteui, int_hexagon_C4_cmplteui, u9ExtPred>;
+def : T_RI_pat<C4_cmpneqi, int_hexagon_C4_cmpneqi, s32ImmPred>;
+def : T_RI_pat<C4_cmpltei, int_hexagon_C4_cmpltei, s32ImmPred>;
+def : T_RI_pat<C4_cmplteui, int_hexagon_C4_cmplteui, u32ImmPred>;
 
 def: T_RR_pat<A4_rcmpeq,  int_hexagon_A4_rcmpeq>;
 def: T_RR_pat<A4_rcmpneq, int_hexagon_A4_rcmpneq>;