From 0e4d7ba5f42c0649d9dcc905d29b04d5a11ecd9c Mon Sep 17 00:00:00 2001 From: notaz Date: Sat, 8 Mar 2008 20:44:21 +0000 Subject: [PATCH] svp compiler: all ops implemented, EXT regs left git-svn-id: file:///home/notaz/opt/svn/PicoDrive@376 be3aeb3a-fb24-0410-a615-afba39da0efa --- Pico/carthw/svp/compiler.c | 306 +++++++++++++++++++++++++++++-------- Pico/carthw/svp/gen_arm.c | 1 + 2 files changed, 241 insertions(+), 66 deletions(-) diff --git a/Pico/carthw/svp/compiler.c b/Pico/carthw/svp/compiler.c index 527647e..fcef34a 100644 --- a/Pico/carthw/svp/compiler.c +++ b/Pico/carthw/svp/compiler.c @@ -12,7 +12,7 @@ static int nblocks = 0; static int iram_context = 0; #ifndef ARM -#define DUMP_BLOCK 0x341e +#define DUMP_BLOCK 0x08aa unsigned int tcache[512*1024]; void regfile_load(void){} void regfile_store(void){} @@ -590,16 +590,17 @@ static void tr_unhandled(void) exit(1); } -/* update P, if needed. Trashes r1 */ +/* update P, if needed. Trashes r0 */ static void tr_flush_dirty_P(void) { // TODO: const regs if (!(dirty_regb & KRREG_P)) return; EOP_MOV_REG_ASR(10, 4, 16); // mov r10, r4, asr #16 - EOP_MOV_REG_LSL( 1, 4, 16); // mov r1, r4, lsl #16 - EOP_MOV_REG_ASR( 1, 1, 15); // mov r1, r1, asr #15 - EOP_MUL(10, 1, 10); // mul r10, r1, r10 + EOP_MOV_REG_LSL( 0, 4, 16); // mov r0, r4, lsl #16 + EOP_MOV_REG_ASR( 0, 0, 15); // mov r0, r0, asr #15 + EOP_MUL(10, 0, 10); // mul r10, r0, r10 dirty_regb &= ~KRREG_P; + hostreg_r[0] = -1; } /* write dirty pr to host reg. Nothing is trashed */ @@ -693,7 +694,7 @@ static void tr_mov16_cond(int cond, int r, int val) hostreg_r[r] = -1; } -/* read bank word to r0. Thrashes r1. */ +/* read bank word to r0 (upper bits zero). Thrashes r1. */ static void tr_bank_read(int addr) /* word addr 0-0x1ff */ { int breg = 7; @@ -740,9 +741,16 @@ static void tr_ptrr_mod(int r, int mod, int need_modulo, int count) { int reg = (r < 4) ? 8 : 9; tr_release_pr(r); - tr_flush_dirty_ST(); - EOP_C_DOP_IMM(A_COND_AL,A_OP_AND,1,6,1,0,0x70); // ands r1, r6, #0x70 - EOP_C_DOP_IMM(A_COND_EQ,A_OP_MOV,0,0,1,0,0x80); // moveq r1, #0x80 + if (dirty_regb & KRREG_ST) { + // avoid flushing ARM flags + EOP_AND_IMM(1, 6, 0, 0x70); + EOP_SUB_IMM(1, 1, 0, 0x10); + EOP_AND_IMM(1, 1, 0, 0x70); + EOP_ADD_IMM(1, 1, 0, 0x10); + } else { + EOP_C_DOP_IMM(A_COND_AL,A_OP_AND,1,6,1,0,0x70); // ands r1, r6, #0x70 + EOP_C_DOP_IMM(A_COND_EQ,A_OP_MOV,0,0,1,0,0x80); // moveq r1, #0x80 + } EOP_MOV_REG_LSR(1, 1, 4); // mov r1, r1, lsr #4 EOP_RSB_IMM(2, 1, 0, 8); // rsb r1, r1, #8 EOP_MOV_IMM(3, 8/2, count); // mov r3, #0x01000000 @@ -776,7 +784,7 @@ static void tr_ptrr_mod(int r, int mod, int need_modulo, int count) /* handle writes r0 to (rX). Trashes r1. * fortunately we can ignore modulo increment modes for writes. */ -static void tr_rX_write1(int op) +static void tr_rX_write(int op) { if ((op&3) == 3) { @@ -823,12 +831,45 @@ static void tr_rX_read(int r, int mod) if (r&3) EOP_ADD_REG_LSR(1,7,1, (r&3)*8-1); // add r1, r7, r1, lsr #lsr else EOP_ADD_REG_LSL(1,7,1,1); EOP_LDRH_SIMPLE(0,1); // ldrh r0, [r1] - hostreg_r[1] = -1; + hostreg_r[0] = hostreg_r[1] = -1; } tr_ptrr_mod(r, mod, 1, 1); } } +/* read ((rX)) to r0. Trashes r1,r2. */ +static void tr_rX_read2(int op) +{ + int r = (op&3) | ((op>>6)&4); // src + + if ((r&3) == 3) { + tr_bank_read((op&0x100) | ((op>>2)&3)); + } else if (known_regb & (1 << (r+8))) { + tr_bank_read((op&0x100) | known_regs.r[r]); + } else { + int reg = (r < 4) ? 8 : 9; + int ror = ((4 - (r&3))*8) & 0x1f; + EOP_AND_IMM(1,reg,ror/2,0xff); // and r1, r{7,8}, + if (r >= 4) + EOP_ORR_IMM(1,1,((ror-8)&0x1f)/2,1); // orr r1, r1, 1<>2)&3)); + } else if (known_regb & (1 << (r+8))) { + tr_bank_write((op&0x100) | known_regs.r[r]); + } else { + EOP_STRH_SIMPLE(0,1); // strh r0, [r1] + hostreg_r[1] = -1; + } + EOP_LDRH_SIMPLE(0,2); // ldrh r0, [r2] + hostreg_r[0] = hostreg_r[2] = -1; +} /* get ARM cond which would mean that SSP cond is satisfied. No trash. */ static int tr_cond_check(int op) @@ -1036,10 +1077,37 @@ static tr_write_func *tr_write_funcs[8] = (tr_write_func *)tr_unhandled }; +static void tr_mac_load_XY(int op) +{ + tr_rX_read(op&3, (op>>2)&3); // X + EOP_MOV_REG_LSL(4, 0, 16); + tr_rX_read(((op>>4)&3)|4, (op>>6)&3); // Y + EOP_ORR_REG_SIMPLE(4, 0); + dirty_regb |= KRREG_P; + hostreg_sspreg_changed(SSP_X); + hostreg_sspreg_changed(SSP_Y); + known_regb &= ~KRREG_X; + known_regb &= ~KRREG_Y; +} + +static int tr_aop_ssp2arm(int op) +{ + switch (op) { + case 1: return A_OP_SUB; + case 3: return A_OP_CMP; + case 4: return A_OP_ADD; + case 5: return A_OP_AND; + case 6: return A_OP_ORR; + case 7: return A_OP_EOR; + } + + tr_unhandled(); + return 0; +} static int translate_op(unsigned int op, int *pc, int imm) { - u32 tmpv, tmpv2; + u32 tmpv, tmpv2, tmpv3; int ret = 0; known_regs.gr[SSP_PC].h = *pc; @@ -1081,7 +1149,7 @@ static int translate_op(unsigned int op, int *pc, int imm) tmpv = (op >> 4) & 0xf; // src if (tmpv >= 8) return -1; // TODO tr_read_funcs[tmpv](); - tr_rX_write1(op); + tr_rX_write(op); ret++; break; // ld a, adr @@ -1129,47 +1197,17 @@ static int translate_op(unsigned int op, int *pc, int imm) return -1; /* TODO.. */ // ld d, ((ri)) - case 0x05: { - int r; - r = (op&3) | ((op>>6)&4); // src + case 0x05: tmpv2 = (op >> 4) & 0xf; // dst if (tmpv2 >= 8) return -1; // TODO - - if ((r&3) == 3) { - tr_bank_read((op&0x100) | ((op>>2)&3)); - } else if (known_regb & (1 << (r+8))) { - tr_bank_read((op&0x100) | known_regs.r[r]); - } else { - int reg = (r < 4) ? 8 : 9; - int ror = ((4 - (r&3))*8) & 0x1f; - EOP_AND_IMM(1,reg,ror/2,0xff); // and r1, r{7,8}, - if (r >= 4) - EOP_ORR_IMM(1,1,((ror-8)&0x1f)/2,1); // orr r1, r1, 1<>2)&3)); - } else if (known_regb & (1 << (r+8))) { - tr_bank_write((op&0x100) | known_regs.r[r]); - } else { - EOP_STRH_SIMPLE(0,1); // strh r0, [r1] - hostreg_r[1] = -1; - } - EOP_LDRH_SIMPLE(0,2); // ldrh r0, [r2] - hostreg_r[0] = hostreg_r[2] = -1; + tr_rX_read2(op); tr_write_funcs[tmpv2](-1); - ret += 3; break; /* should certainly take > 1 */ - } + ret += 3; break; // ldi (ri), imm case 0x06: tr_mov16(0, imm); - tr_rX_write1(op); + tr_rX_write(op); ret += 2; break; // ld adr, a @@ -1307,35 +1345,171 @@ static int translate_op(unsigned int op, int *pc, int imm) known_regb &= ~(KRREG_A|KRREG_AL); ret += tmpv; break; } -/* + // mpys? case 0x1b: - read_P(); // update P - rA32 -= rP.v; // maybe only upper word? - UPD_ACC_ZN // there checking flags after this - rX = ptr1_read_(op&3, 0, (op<<1)&0x18); // ri (maybe rj?) - rY = ptr1_read_((op>>4)&3, 4, (op>>3)&0x18); // rj - break; + tr_flush_dirty_P(); + tr_mac_load_XY(op); + tr_make_dirty_ST(); + EOP_C_DOP_REG_XIMM(A_COND_AL,A_OP_SUB,1,5,5,0,A_AM1_LSL,10); // subs r5, r5, r10 + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL); + dirty_regb |= KRREG_ST; + ret++; break; // mpya (rj), (ri), b case 0x4b: - read_P(); // update P - rA32 += rP.v; // confirmed to be 32bit - UPD_ACC_ZN // ? - rX = ptr1_read_(op&3, 0, (op<<1)&0x18); // ri (maybe rj?) - rY = ptr1_read_((op>>4)&3, 4, (op>>3)&0x18); // rj - break; + tr_flush_dirty_P(); + tr_mac_load_XY(op); + tr_make_dirty_ST(); + EOP_C_DOP_REG_XIMM(A_COND_AL,A_OP_ADD,1,5,5,0,A_AM1_LSL,10); // adds r5, r5, r10 + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL); + dirty_regb |= KRREG_ST; + ret++; break; // mld (rj), (ri), b case 0x5b: - EOP_MOV_IMM(5, 0, 0); // mov r5, #0 - known_regs.r[SSP_A].v = 0; + EOP_C_DOP_IMM(A_COND_AL,A_OP_MOV,1,0,5,0,0); // movs r5, #0 + hostreg_sspreg_changed(SSP_A); + known_regs.gr[SSP_A].v = 0; known_regb |= (KRREG_A|KRREG_AL); - EOP_BIC_IMM(6, 6, 0, 0x0f); // bic r6, r6, 0xf // flags - EOP_BIC_IMM(6, 6, 0, 0x04); // bic r6, r6, 4 // set Z - // TODO + dirty_regb |= KRREG_ST; + tr_mac_load_XY(op); + ret++; break; + + // OP a, s + case 0x10: + case 0x30: + case 0x40: + case 0x50: + case 0x60: + case 0x70: + tmpv = op & 0xf; // src + tmpv2 = tr_aop_ssp2arm(op>>13); // op + tmpv3 = (tmpv2 == A_OP_CMP) ? 0 : 5; + if (tmpv >= 8) return -1; // TODO + if (tmpv == SSP_P) { + tr_flush_dirty_P(); + EOP_C_DOP_REG_XIMM(A_COND_AL,tmpv2,1,5,tmpv3, 0,A_AM1_LSL,10); // OPs r5, r5, r10 + } else if (tmpv == SSP_A) { + EOP_C_DOP_REG_XIMM(A_COND_AL,tmpv2,1,5,tmpv3, 0,A_AM1_LSL, 5); // OPs r5, r5, r5 + } else { + tr_read_funcs[tmpv](); + EOP_C_DOP_REG_XIMM(A_COND_AL,tmpv2,1,5,tmpv3,16,A_AM1_LSL, 0); // OPs r5, r5, r0, lsl #16 + } + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL|KRREG_ST); + dirty_regb |= KRREG_ST; + ret++; break; + + // OP a, (ri) + case 0x11: + case 0x31: + case 0x41: + case 0x51: + case 0x61: + case 0x71: + tmpv2 = tr_aop_ssp2arm(op>>13); // op + tmpv3 = (tmpv2 == A_OP_CMP) ? 0 : 5; + tr_rX_read((op&3)|((op>>6)&4), (op>>2)&3); + EOP_C_DOP_REG_XIMM(A_COND_AL,tmpv2,1,5,tmpv3,16,A_AM1_LSL,0); // OPs r5, r5, r0, lsl #16 + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL|KRREG_ST); + dirty_regb |= KRREG_ST; + ret++; break; + + // OP a, adr + case 0x13: + case 0x33: + case 0x43: + case 0x53: + case 0x63: + case 0x73: + tmpv2 = tr_aop_ssp2arm(op>>13); // op + tmpv3 = (tmpv2 == A_OP_CMP) ? 0 : 5; + tr_bank_read(op&0x1ff); + EOP_C_DOP_REG_XIMM(A_COND_AL,tmpv2,1,5,tmpv3,16,A_AM1_LSL,0); // OPs r5, r5, r0, lsl #16 + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL|KRREG_ST); + dirty_regb |= KRREG_ST; + ret++; break; + + // OP a, imm + case 0x14: + case 0x34: + case 0x44: + case 0x54: + case 0x64: + case 0x74: + tmpv = (op & 0xf0) >> 4; + tmpv2 = tr_aop_ssp2arm(op>>13); // op + tmpv3 = (tmpv2 == A_OP_CMP) ? 0 : 5; + tr_mov16(0, imm); + EOP_C_DOP_REG_XIMM(A_COND_AL,tmpv2,1,5,tmpv3,16,A_AM1_LSL,0); // OPs r5, r5, r0, lsl #16 + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL|KRREG_ST); + dirty_regb |= KRREG_ST; + ret += 2; break; + + // OP a, ((ri)) + case 0x15: + case 0x35: + case 0x45: + case 0x55: + case 0x65: + case 0x75: + tmpv2 = tr_aop_ssp2arm(op>>13); // op + tmpv3 = (tmpv2 == A_OP_CMP) ? 0 : 5; + tr_rX_read2(op); + EOP_C_DOP_REG_XIMM(A_COND_AL,tmpv2,1,5,tmpv3,16,A_AM1_LSL,0); // OPs r5, r5, r0, lsl #16 + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL|KRREG_ST); + dirty_regb |= KRREG_ST; + ret += 3; break; + + // OP a, ri + case 0x19: + case 0x39: + case 0x49: + case 0x59: + case 0x69: + case 0x79: { + int r; + tmpv2 = tr_aop_ssp2arm(op>>13); // op + tmpv3 = (tmpv2 == A_OP_CMP) ? 0 : 5; + r = (op&3) | ((op>>6)&4); // src + if ((r&3) == 3) tr_unhandled(); + + if (known_regb & (1 << (r+8))) { + EOP_C_DOP_IMM(A_COND_AL,tmpv2,1,5,tmpv3,16/2,known_regs.r[r]); // OPs r5, r5, #val<<16 + } else { + int reg = (r < 4) ? 8 : 9; + if (r&3) EOP_MOV_REG_LSR(0, reg, (r&3)*8); // mov r0, r{7,8}, lsr #lsr + EOP_AND_IMM(0, (r&3)?0:reg, 0, 0xff); // and r0, r{7,8}, + EOP_C_DOP_REG_XIMM(A_COND_AL,tmpv2,1,5,tmpv3,16,A_AM1_LSL,0); // OPs r5, r5, r0, lsl #16 + hostreg_r[0] = -1; + } + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL|KRREG_ST); + dirty_regb |= KRREG_ST; + ret++; break; + } + + // OP simm + case 0x1c: + case 0x3c: + case 0x4c: + case 0x5c: + case 0x6c: + case 0x7c: + tmpv2 = tr_aop_ssp2arm(op>>13); // op + tmpv3 = (tmpv2 == A_OP_CMP) ? 0 : 5; + EOP_C_DOP_IMM(A_COND_AL,tmpv2,1,5,tmpv3,16/2,op & 0xff); // OPs r5, r5, #val<<16 + hostreg_sspreg_changed(SSP_A); + known_regb &= ~(KRREG_A|KRREG_AL|KRREG_ST); + dirty_regb |= KRREG_ST; ret++; break; -*/ } return ret; diff --git a/Pico/carthw/svp/gen_arm.c b/Pico/carthw/svp/gen_arm.c index 4e8f05b..87e8368 100644 --- a/Pico/carthw/svp/gen_arm.c +++ b/Pico/carthw/svp/gen_arm.c @@ -33,6 +33,7 @@ #define A_OP_RSB 0x3 #define A_OP_ADD 0x4 #define A_OP_TST 0x8 +#define A_OP_CMP 0xa #define A_OP_ORR 0xc #define A_OP_MOV 0xd #define A_OP_BIC 0xe -- 2.39.2