| 1 | ! |
|---|
| 2 | ! Fast SH memset |
|---|
| 3 | ! |
|---|
| 4 | ! by Toshiyasu Morita (tm@netcom.com) |
|---|
| 5 | ! |
|---|
| 6 | ! SH5 code by J"orn Rennecke (joern.rennecke@superh.com) |
|---|
| 7 | ! Copyright 2002 SuperH Ltd. |
|---|
| 8 | ! |
|---|
| 9 | |
|---|
| 10 | #include "asm.h" |
|---|
| 11 | |
|---|
| 12 | ENTRY(memset) |
|---|
| 13 | #if __SHMEDIA__ |
|---|
| 14 | pta/l multiquad, tr0 |
|---|
| 15 | ptabs r18, tr2 |
|---|
| 16 | |
|---|
| 17 | andi r2, -8, r25 |
|---|
| 18 | add r2, r4, r5 |
|---|
| 19 | addi r5, -1, r20 // calculate end address. |
|---|
| 20 | andi r20, -8, r20 |
|---|
| 21 | cmveq r4, r25, r20 |
|---|
| 22 | bne/u r25, r20, tr0 // multiquad |
|---|
| 23 | |
|---|
| 24 | ! This sequence could clobber volatile objects that are in the same |
|---|
| 25 | ! quadword as a very short char array. |
|---|
| 26 | ! ldlo.q r2, 0, r7 |
|---|
| 27 | ! shlli r4, 2, r4 |
|---|
| 28 | ! movi -1, r8 |
|---|
| 29 | ! SHHI r8, r4, r8 |
|---|
| 30 | ! SHHI r8, r4, r8 |
|---|
| 31 | ! mcmv r7, r8, r3 |
|---|
| 32 | ! stlo.q r2, 0, r3 |
|---|
| 33 | |
|---|
| 34 | pta/l setlongs, tr0 |
|---|
| 35 | movi 4, r8 |
|---|
| 36 | bgeu/u r4, r8, tr0 |
|---|
| 37 | pta/l endset, tr0 |
|---|
| 38 | beqi/u r4, 0, tr0 |
|---|
| 39 | st.b r2, 0, r3 |
|---|
| 40 | beqi/u r4, 1, tr0 |
|---|
| 41 | nop |
|---|
| 42 | st.b r2, 1, r3 |
|---|
| 43 | beqi/l r4, 2, tr0 |
|---|
| 44 | st.b r2,2,r3 |
|---|
| 45 | endset: blink tr2, r63 |
|---|
| 46 | setlongs: |
|---|
| 47 | mshflo.b r3, r3, r3 |
|---|
| 48 | mperm.w r3, r63, r3 // Fill pattern now in every byte of r3 |
|---|
| 49 | stlo.l r2, 0, r3 |
|---|
| 50 | nop |
|---|
| 51 | nop |
|---|
| 52 | sthi.l r5, -1, r3 |
|---|
| 53 | blink tr2, r63 |
|---|
| 54 | |
|---|
| 55 | multiquad: |
|---|
| 56 | mshflo.b r3, r3, r3 |
|---|
| 57 | mperm.w r3, r63, r3 // Fill pattern now in every byte of r3 |
|---|
| 58 | pta/l lastquad, tr0 |
|---|
| 59 | stlo.q r2, 0, r3 |
|---|
| 60 | sub r20, r25, r24 |
|---|
| 61 | movi 64, r9 |
|---|
| 62 | beqi/u r24, 8, tr0 // lastquad |
|---|
| 63 | pta/l loop, tr1 |
|---|
| 64 | addi r20, -7*8, r8 // loop end address; This might overflow, so we need |
|---|
| 65 | // to use a different test before we start the loop |
|---|
| 66 | bgeu/u r24, r9, tr1// loop |
|---|
| 67 | st.q r25, 8, r3 |
|---|
| 68 | shlri r24, 4, r24 |
|---|
| 69 | st.q r20, -8, r3 |
|---|
| 70 | beqi/u r24, 1, tr0 // lastquad |
|---|
| 71 | st.q r25, 16, r3 |
|---|
| 72 | st.q r20, -16, r3 |
|---|
| 73 | beqi/u r24, 2, tr0 // lastquad |
|---|
| 74 | st.q r25, 24, r3 |
|---|
| 75 | st.q r20, -24, r3 |
|---|
| 76 | lastquad: |
|---|
| 77 | sthi.q r5, -1, r3 |
|---|
| 78 | blink tr2,r63 |
|---|
| 79 | |
|---|
| 80 | loop: |
|---|
| 81 | alloco r25, 32 |
|---|
| 82 | st.q r25, 8, r3 |
|---|
| 83 | st.q r25, 16, r3 |
|---|
| 84 | st.q r25, 24, r3 |
|---|
| 85 | st.q r25, 32, r3 |
|---|
| 86 | addi r25, 32, r25 |
|---|
| 87 | bgeu/l r8, r25, tr1 // loop |
|---|
| 88 | |
|---|
| 89 | st.q r20, -40, r3 |
|---|
| 90 | st.q r20, -32, r3 |
|---|
| 91 | st.q r20, -24, r3 |
|---|
| 92 | st.q r20, -16, r3 |
|---|
| 93 | st.q r20, -8, r3 |
|---|
| 94 | sthi.q r5, -1, r3 |
|---|
| 95 | blink tr2,r63 |
|---|
| 96 | #else /* ! SHMEDIA, i.e. SH1 .. SH4 / SHcompact */ |
|---|
| 97 | ! Entry: r4: destination pointer |
|---|
| 98 | ! r5: fill value |
|---|
| 99 | ! r6: byte count |
|---|
| 100 | ! |
|---|
| 101 | ! Exit: r0-r3: trashed |
|---|
| 102 | ! |
|---|
| 103 | |
|---|
| 104 | ! This assumes that the first four bytes of the address space (0..3) are |
|---|
| 105 | ! reserved - usually by the linker script. Otherwise, we would had to check |
|---|
| 106 | ! for the case of objects of the size 12..15 at address 0..3 . |
|---|
| 107 | |
|---|
| 108 | #ifdef __SH5__ |
|---|
| 109 | #define DST r2 |
|---|
| 110 | #define VAL r3 |
|---|
| 111 | #define CNT r4 |
|---|
| 112 | #define TMP r5 |
|---|
| 113 | #else |
|---|
| 114 | #define DST r4 |
|---|
| 115 | #define VAL r5 |
|---|
| 116 | #define CNT r6 |
|---|
| 117 | #define TMP r2 |
|---|
| 118 | #endif |
|---|
| 119 | |
|---|
| 120 | mov #12,r0 ! Check for small number of bytes |
|---|
| 121 | cmp/gt CNT,r0 |
|---|
| 122 | mov DST,r0 |
|---|
| 123 | SL(bt, L_store_byte_loop_check0, add DST,CNT) |
|---|
| 124 | |
|---|
| 125 | tst #3,r0 ! Align destination |
|---|
| 126 | SL(bt, L_dup_bytes, extu.b r5,r5) |
|---|
| 127 | .balignw 4,0x0009 |
|---|
| 128 | L_align_loop: |
|---|
| 129 | mov.b VAL,@r0 |
|---|
| 130 | add #1,r0 |
|---|
| 131 | tst #3,r0 |
|---|
| 132 | bf L_align_loop |
|---|
| 133 | |
|---|
| 134 | L_dup_bytes: |
|---|
| 135 | swap.b VAL,TMP ! Duplicate bytes across longword |
|---|
| 136 | or TMP,VAL |
|---|
| 137 | swap.w VAL,TMP |
|---|
| 138 | or TMP,VAL |
|---|
| 139 | |
|---|
| 140 | add #-16,CNT |
|---|
| 141 | |
|---|
| 142 | .balignw 4,0x0009 |
|---|
| 143 | L_store_long_loop: |
|---|
| 144 | mov.l VAL,@r0 ! Store double longs to memory |
|---|
| 145 | cmp/hs CNT,r0 |
|---|
| 146 | mov.l VAL,@(4,r0) |
|---|
| 147 | SL(bf, L_store_long_loop, add #8,r0) |
|---|
| 148 | |
|---|
| 149 | add #16,CNT |
|---|
| 150 | |
|---|
| 151 | L_store_byte_loop_check0: |
|---|
| 152 | cmp/eq CNT,r0 |
|---|
| 153 | bt L_exit |
|---|
| 154 | .balignw 4,0x0009 |
|---|
| 155 | L_store_byte_loop: |
|---|
| 156 | mov.b VAL,@r0 ! Store bytes to memory |
|---|
| 157 | add #1,r0 |
|---|
| 158 | cmp/eq CNT,r0 |
|---|
| 159 | bf L_store_byte_loop |
|---|
| 160 | |
|---|
| 161 | L_exit: |
|---|
| 162 | rts |
|---|
| 163 | mov r4,r0 |
|---|
| 164 | #endif /* ! SHMEDIA */ |
|---|
