From f515d553f0a92081e4eb2fa84ef0f62e9614d14e Mon Sep 17 00:00:00 2001 From: zherczeg Date: Fri, 4 Mar 2016 08:51:53 +0000 Subject: JIT compiler update. git-svn-id: svn://vcs.exim.org/pcre2/code/trunk@505 6239d852-aaf2-0410-a92c-79f79f948069 --- src/sljit/sljitNativeX86_common.c | 144 +++++++++++++++++++------------------- 1 file changed, 72 insertions(+), 72 deletions(-) (limited to 'src/sljit/sljitNativeX86_common.c') diff --git a/src/sljit/sljitNativeX86_common.c b/src/sljit/sljitNativeX86_common.c index 9062fe0..54c3ac7 100644 --- a/src/sljit/sljitNativeX86_common.c +++ b/src/sljit/sljitNativeX86_common.c @@ -334,27 +334,27 @@ static sljit_u8 get_jump_code(sljit_s32 type) { switch (type) { case SLJIT_EQUAL: - case SLJIT_D_EQUAL: + case SLJIT_EQUAL_F64: return 0x84 /* je */; case SLJIT_NOT_EQUAL: - case SLJIT_D_NOT_EQUAL: + case SLJIT_NOT_EQUAL_F64: return 0x85 /* jne */; case SLJIT_LESS: - case SLJIT_D_LESS: + case SLJIT_LESS_F64: return 0x82 /* jc */; case SLJIT_GREATER_EQUAL: - case SLJIT_D_GREATER_EQUAL: + case SLJIT_GREATER_EQUAL_F64: return 0x83 /* jae */; case SLJIT_GREATER: - case SLJIT_D_GREATER: + case SLJIT_GREATER_F64: return 0x87 /* jnbe */; case SLJIT_LESS_EQUAL: - case SLJIT_D_LESS_EQUAL: + case SLJIT_LESS_EQUAL_F64: return 0x86 /* jbe */; case SLJIT_SIG_LESS: @@ -377,10 +377,10 @@ static sljit_u8 get_jump_code(sljit_s32 type) case SLJIT_MUL_NOT_OVERFLOW: return 0x81 /* jno */; - case SLJIT_D_UNORDERED: + case SLJIT_UNORDERED_F64: return 0x8a /* jp */; - case SLJIT_D_ORDERED: + case SLJIT_ORDERED_F64: return 0x8b /* jpo */; } return 0; @@ -742,12 +742,12 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_op0(struct sljit_compiler *compile INC_SIZE(1); *inst = NOP; break; - case SLJIT_LUMUL: - case SLJIT_LSMUL: - case SLJIT_UDIVMOD: - case SLJIT_SDIVMOD: - case SLJIT_UDIVI: - case SLJIT_SDIVI: + case SLJIT_LMUL_UW: + case SLJIT_LMUL_SW: + case SLJIT_DIVMOD_UW: + case SLJIT_DIVMOD_SW: + case SLJIT_DIV_UW: + case SLJIT_DIV_SW: compiler->flags_saved = 0; #if (defined SLJIT_CONFIG_X86_64 && SLJIT_CONFIG_X86_64) #ifdef _WIN64 @@ -765,10 +765,10 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_op0(struct sljit_compiler *compile #endif compiler->mode32 = op & SLJIT_I32_OP; #endif - SLJIT_COMPILE_ASSERT((SLJIT_UDIVMOD & 0x2) == 0 && SLJIT_UDIVI - 0x2 == SLJIT_UDIVMOD, bad_div_opcode_assignments); + SLJIT_COMPILE_ASSERT((SLJIT_DIVMOD_UW & 0x2) == 0 && SLJIT_DIV_UW - 0x2 == SLJIT_DIVMOD_UW, bad_div_opcode_assignments); op = GET_OPCODE(op); - if ((op | 0x2) == SLJIT_UDIVI) { + if ((op | 0x2) == SLJIT_DIV_UW) { #if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32) || defined(_WIN64) EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_R1, 0); inst = emit_x86_instruction(compiler, 1, SLJIT_R1, 0, SLJIT_R1, 0); @@ -779,7 +779,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_op0(struct sljit_compiler *compile *inst = XOR_r_rm; } - if ((op | 0x2) == SLJIT_SDIVI) { + if ((op | 0x2) == SLJIT_DIV_SW) { #if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32) || defined(_WIN64) EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_R1, 0); #endif @@ -810,10 +810,10 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_op0(struct sljit_compiler *compile FAIL_IF(!inst); INC_SIZE(2); *inst++ = GROUP_F7; - *inst = MOD_REG | ((op >= SLJIT_UDIVMOD) ? reg_map[TMP_REG1] : reg_map[SLJIT_R1]); + *inst = MOD_REG | ((op >= SLJIT_DIVMOD_UW) ? reg_map[TMP_REG1] : reg_map[SLJIT_R1]); #else #ifdef _WIN64 - size = (!compiler->mode32 || op >= SLJIT_UDIVMOD) ? 3 : 2; + size = (!compiler->mode32 || op >= SLJIT_DIVMOD_UW) ? 3 : 2; #else size = (!compiler->mode32) ? 3 : 2; #endif @@ -822,11 +822,11 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_op0(struct sljit_compiler *compile INC_SIZE(size); #ifdef _WIN64 if (!compiler->mode32) - *inst++ = REX_W | ((op >= SLJIT_UDIVMOD) ? REX_B : 0); - else if (op >= SLJIT_UDIVMOD) + *inst++ = REX_W | ((op >= SLJIT_DIVMOD_UW) ? REX_B : 0); + else if (op >= SLJIT_DIVMOD_UW) *inst++ = REX_B; *inst++ = GROUP_F7; - *inst = MOD_REG | ((op >= SLJIT_UDIVMOD) ? reg_lmap[TMP_REG1] : reg_lmap[SLJIT_R1]); + *inst = MOD_REG | ((op >= SLJIT_DIVMOD_UW) ? reg_lmap[TMP_REG1] : reg_lmap[SLJIT_R1]); #else if (!compiler->mode32) *inst++ = REX_W; @@ -835,26 +835,26 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_op0(struct sljit_compiler *compile #endif #endif switch (op) { - case SLJIT_LUMUL: + case SLJIT_LMUL_UW: *inst |= MUL; break; - case SLJIT_LSMUL: + case SLJIT_LMUL_SW: *inst |= IMUL; break; - case SLJIT_UDIVMOD: - case SLJIT_UDIVI: + case SLJIT_DIVMOD_UW: + case SLJIT_DIV_UW: *inst |= DIV; break; - case SLJIT_SDIVMOD: - case SLJIT_SDIVI: + case SLJIT_DIVMOD_SW: + case SLJIT_DIV_SW: *inst |= IDIV; break; } #if (defined SLJIT_CONFIG_X86_64 && SLJIT_CONFIG_X86_64) && !defined(_WIN64) - if (op <= SLJIT_SDIVMOD) + if (op <= SLJIT_DIVMOD_SW) EMIT_MOV(compiler, SLJIT_R1, 0, TMP_REG1, 0); #else - if (op >= SLJIT_UDIVI) + if (op >= SLJIT_DIV_UW) EMIT_MOV(compiler, SLJIT_R1, 0, TMP_REG1, 0); #endif break; @@ -2322,7 +2322,7 @@ static SLJIT_INLINE sljit_s32 emit_sse2_store(struct sljit_compiler *compiler, return emit_sse2(compiler, MOVSD_xm_x, single, src, dst, dstw); } -static SLJIT_INLINE sljit_s32 sljit_emit_fop1_convw_fromd(struct sljit_compiler *compiler, sljit_s32 op, +static SLJIT_INLINE sljit_s32 sljit_emit_fop1_conv_sw_from_f64(struct sljit_compiler *compiler, sljit_s32 op, sljit_s32 dst, sljit_sw dstw, sljit_s32 src, sljit_sw srcw) { @@ -2330,11 +2330,11 @@ static SLJIT_INLINE sljit_s32 sljit_emit_fop1_convw_fromd(struct sljit_compiler sljit_u8 *inst; #if (defined SLJIT_CONFIG_X86_64 && SLJIT_CONFIG_X86_64) - if (GET_OPCODE(op) == SLJIT_CONVW_FROMD) + if (GET_OPCODE(op) == SLJIT_CONV_SW_FROM_F64) compiler->mode32 = 0; #endif - inst = emit_x86_instruction(compiler, 2 | ((op & SLJIT_SINGLE_OP) ? EX86_PREF_F3 : EX86_PREF_F2) | EX86_SSE2_OP2, dst_r, 0, src, srcw); + inst = emit_x86_instruction(compiler, 2 | ((op & SLJIT_F32_OP) ? EX86_PREF_F3 : EX86_PREF_F2) | EX86_SSE2_OP2, dst_r, 0, src, srcw); FAIL_IF(!inst); *inst++ = GROUP_0F; *inst = CVTTSD2SI_r_xm; @@ -2344,7 +2344,7 @@ static SLJIT_INLINE sljit_s32 sljit_emit_fop1_convw_fromd(struct sljit_compiler return SLJIT_SUCCESS; } -static SLJIT_INLINE sljit_s32 sljit_emit_fop1_convd_fromw(struct sljit_compiler *compiler, sljit_s32 op, +static SLJIT_INLINE sljit_s32 sljit_emit_fop1_conv_f64_from_sw(struct sljit_compiler *compiler, sljit_s32 op, sljit_s32 dst, sljit_sw dstw, sljit_s32 src, sljit_sw srcw) { @@ -2352,13 +2352,13 @@ static SLJIT_INLINE sljit_s32 sljit_emit_fop1_convd_fromw(struct sljit_compiler sljit_u8 *inst; #if (defined SLJIT_CONFIG_X86_64 && SLJIT_CONFIG_X86_64) - if (GET_OPCODE(op) == SLJIT_CONVD_FROMW) + if (GET_OPCODE(op) == SLJIT_CONV_F64_FROM_SW) compiler->mode32 = 0; #endif if (src & SLJIT_IMM) { #if (defined SLJIT_CONFIG_X86_64 && SLJIT_CONFIG_X86_64) - if (GET_OPCODE(op) == SLJIT_CONVD_FROMI) + if (GET_OPCODE(op) == SLJIT_CONV_F64_FROM_S32) srcw = (sljit_s32)srcw; #endif EMIT_MOV(compiler, TMP_REG1, 0, src, srcw); @@ -2366,7 +2366,7 @@ static SLJIT_INLINE sljit_s32 sljit_emit_fop1_convd_fromw(struct sljit_compiler srcw = 0; } - inst = emit_x86_instruction(compiler, 2 | ((op & SLJIT_SINGLE_OP) ? EX86_PREF_F3 : EX86_PREF_F2) | EX86_SSE2_OP1, dst_r, 0, src, srcw); + inst = emit_x86_instruction(compiler, 2 | ((op & SLJIT_F32_OP) ? EX86_PREF_F3 : EX86_PREF_F2) | EX86_SSE2_OP1, dst_r, 0, src, srcw); FAIL_IF(!inst); *inst++ = GROUP_0F; *inst = CVTSI2SD_x_rm; @@ -2375,7 +2375,7 @@ static SLJIT_INLINE sljit_s32 sljit_emit_fop1_convd_fromw(struct sljit_compiler compiler->mode32 = 1; #endif if (dst_r == TMP_FREG) - return emit_sse2_store(compiler, op & SLJIT_SINGLE_OP, dst, dstw, TMP_FREG); + return emit_sse2_store(compiler, op & SLJIT_F32_OP, dst, dstw, TMP_FREG); return SLJIT_SUCCESS; } @@ -2385,10 +2385,10 @@ static SLJIT_INLINE sljit_s32 sljit_emit_fop1_cmp(struct sljit_compiler *compile { compiler->flags_saved = 0; if (!FAST_IS_REG(src1)) { - FAIL_IF(emit_sse2_load(compiler, op & SLJIT_SINGLE_OP, TMP_FREG, src1, src1w)); + FAIL_IF(emit_sse2_load(compiler, op & SLJIT_F32_OP, TMP_FREG, src1, src1w)); src1 = TMP_FREG; } - return emit_sse2_logic(compiler, UCOMISD_x_xm, !(op & SLJIT_SINGLE_OP), src1, src2, src2w); + return emit_sse2_logic(compiler, UCOMISD_x_xm, !(op & SLJIT_F32_OP), src1, src2, src2w); } SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_fop1(struct sljit_compiler *compiler, sljit_s32 op, @@ -2404,56 +2404,56 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_fop1(struct sljit_compiler *compil CHECK_ERROR(); SELECT_FOP1_OPERATION_WITH_CHECKS(compiler, op, dst, dstw, src, srcw); - if (GET_OPCODE(op) == SLJIT_DMOV) { + if (GET_OPCODE(op) == SLJIT_MOV_F64) { if (FAST_IS_REG(dst)) - return emit_sse2_load(compiler, op & SLJIT_SINGLE_OP, dst, src, srcw); + return emit_sse2_load(compiler, op & SLJIT_F32_OP, dst, src, srcw); if (FAST_IS_REG(src)) - return emit_sse2_store(compiler, op & SLJIT_SINGLE_OP, dst, dstw, src); - FAIL_IF(emit_sse2_load(compiler, op & SLJIT_SINGLE_OP, TMP_FREG, src, srcw)); - return emit_sse2_store(compiler, op & SLJIT_SINGLE_OP, dst, dstw, TMP_FREG); + return emit_sse2_store(compiler, op & SLJIT_F32_OP, dst, dstw, src); + FAIL_IF(emit_sse2_load(compiler, op & SLJIT_F32_OP, TMP_FREG, src, srcw)); + return emit_sse2_store(compiler, op & SLJIT_F32_OP, dst, dstw, TMP_FREG); } - if (GET_OPCODE(op) == SLJIT_CONVD_FROMS) { + if (GET_OPCODE(op) == SLJIT_CONV_F64_FROM_F32) { dst_r = FAST_IS_REG(dst) ? dst : TMP_FREG; if (FAST_IS_REG(src)) { /* We overwrite the high bits of source. From SLJIT point of view, this is not an issue. Note: In SSE3, we could also use MOVDDUP and MOVSLDUP. */ - FAIL_IF(emit_sse2_logic(compiler, UNPCKLPD_x_xm, op & SLJIT_SINGLE_OP, src, src, 0)); + FAIL_IF(emit_sse2_logic(compiler, UNPCKLPD_x_xm, op & SLJIT_F32_OP, src, src, 0)); } else { - FAIL_IF(emit_sse2_load(compiler, !(op & SLJIT_SINGLE_OP), TMP_FREG, src, srcw)); + FAIL_IF(emit_sse2_load(compiler, !(op & SLJIT_F32_OP), TMP_FREG, src, srcw)); src = TMP_FREG; } - FAIL_IF(emit_sse2_logic(compiler, CVTPD2PS_x_xm, op & SLJIT_SINGLE_OP, dst_r, src, 0)); + FAIL_IF(emit_sse2_logic(compiler, CVTPD2PS_x_xm, op & SLJIT_F32_OP, dst_r, src, 0)); if (dst_r == TMP_FREG) - return emit_sse2_store(compiler, op & SLJIT_SINGLE_OP, dst, dstw, TMP_FREG); + return emit_sse2_store(compiler, op & SLJIT_F32_OP, dst, dstw, TMP_FREG); return SLJIT_SUCCESS; } if (SLOW_IS_REG(dst)) { dst_r = dst; if (dst != src) - FAIL_IF(emit_sse2_load(compiler, op & SLJIT_SINGLE_OP, dst_r, src, srcw)); + FAIL_IF(emit_sse2_load(compiler, op & SLJIT_F32_OP, dst_r, src, srcw)); } else { dst_r = TMP_FREG; - FAIL_IF(emit_sse2_load(compiler, op & SLJIT_SINGLE_OP, dst_r, src, srcw)); + FAIL_IF(emit_sse2_load(compiler, op & SLJIT_F32_OP, dst_r, src, srcw)); } switch (GET_OPCODE(op)) { - case SLJIT_DNEG: - FAIL_IF(emit_sse2_logic(compiler, XORPD_x_xm, 1, dst_r, SLJIT_MEM0(), (sljit_sw)(op & SLJIT_SINGLE_OP ? sse2_buffer : sse2_buffer + 8))); + case SLJIT_NEG_F64: + FAIL_IF(emit_sse2_logic(compiler, XORPD_x_xm, 1, dst_r, SLJIT_MEM0(), (sljit_sw)(op & SLJIT_F32_OP ? sse2_buffer : sse2_buffer + 8))); break; - case SLJIT_DABS: - FAIL_IF(emit_sse2_logic(compiler, ANDPD_x_xm, 1, dst_r, SLJIT_MEM0(), (sljit_sw)(op & SLJIT_SINGLE_OP ? sse2_buffer + 4 : sse2_buffer + 12))); + case SLJIT_ABS_F64: + FAIL_IF(emit_sse2_logic(compiler, ANDPD_x_xm, 1, dst_r, SLJIT_MEM0(), (sljit_sw)(op & SLJIT_F32_OP ? sse2_buffer + 4 : sse2_buffer + 12))); break; } if (dst_r == TMP_FREG) - return emit_sse2_store(compiler, op & SLJIT_SINGLE_OP, dst, dstw, TMP_FREG); + return emit_sse2_store(compiler, op & SLJIT_F32_OP, dst, dstw, TMP_FREG); return SLJIT_SUCCESS; } @@ -2478,43 +2478,43 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_fop2(struct sljit_compiler *compil dst_r = dst; if (dst == src1) ; /* Do nothing here. */ - else if (dst == src2 && (op == SLJIT_DADD || op == SLJIT_DMUL)) { + else if (dst == src2 && (op == SLJIT_ADD_F64 || op == SLJIT_MUL_F64)) { /* Swap arguments. */ src2 = src1; src2w = src1w; } else if (dst != src2) - FAIL_IF(emit_sse2_load(compiler, op & SLJIT_SINGLE_OP, dst_r, src1, src1w)); + FAIL_IF(emit_sse2_load(compiler, op & SLJIT_F32_OP, dst_r, src1, src1w)); else { dst_r = TMP_FREG; - FAIL_IF(emit_sse2_load(compiler, op & SLJIT_SINGLE_OP, TMP_FREG, src1, src1w)); + FAIL_IF(emit_sse2_load(compiler, op & SLJIT_F32_OP, TMP_FREG, src1, src1w)); } } else { dst_r = TMP_FREG; - FAIL_IF(emit_sse2_load(compiler, op & SLJIT_SINGLE_OP, TMP_FREG, src1, src1w)); + FAIL_IF(emit_sse2_load(compiler, op & SLJIT_F32_OP, TMP_FREG, src1, src1w)); } switch (GET_OPCODE(op)) { - case SLJIT_DADD: - FAIL_IF(emit_sse2(compiler, ADDSD_x_xm, op & SLJIT_SINGLE_OP, dst_r, src2, src2w)); + case SLJIT_ADD_F64: + FAIL_IF(emit_sse2(compiler, ADDSD_x_xm, op & SLJIT_F32_OP, dst_r, src2, src2w)); break; - case SLJIT_DSUB: - FAIL_IF(emit_sse2(compiler, SUBSD_x_xm, op & SLJIT_SINGLE_OP, dst_r, src2, src2w)); + case SLJIT_SUB_F64: + FAIL_IF(emit_sse2(compiler, SUBSD_x_xm, op & SLJIT_F32_OP, dst_r, src2, src2w)); break; - case SLJIT_DMUL: - FAIL_IF(emit_sse2(compiler, MULSD_x_xm, op & SLJIT_SINGLE_OP, dst_r, src2, src2w)); + case SLJIT_MUL_F64: + FAIL_IF(emit_sse2(compiler, MULSD_x_xm, op & SLJIT_F32_OP, dst_r, src2, src2w)); break; - case SLJIT_DDIV: - FAIL_IF(emit_sse2(compiler, DIVSD_x_xm, op & SLJIT_SINGLE_OP, dst_r, src2, src2w)); + case SLJIT_DIV_F64: + FAIL_IF(emit_sse2(compiler, DIVSD_x_xm, op & SLJIT_F32_OP, dst_r, src2, src2w)); break; } if (dst_r == TMP_FREG) - return emit_sse2_store(compiler, op & SLJIT_SINGLE_OP, dst, dstw, TMP_FREG); + return emit_sse2_store(compiler, op & SLJIT_F32_OP, dst, dstw, TMP_FREG); return SLJIT_SUCCESS; } @@ -2966,7 +2966,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_x86_emit_cmov(struct sljit_compiler *co #if (defined SLJIT_ARGUMENT_CHECKS && SLJIT_ARGUMENT_CHECKS) CHECK_ARGUMENT(sljit_x86_is_cmov_available()); CHECK_ARGUMENT(!(type & ~(0xff | SLJIT_I32_OP))); - CHECK_ARGUMENT((type & 0xff) >= SLJIT_EQUAL && (type & 0xff) <= SLJIT_D_ORDERED); + CHECK_ARGUMENT((type & 0xff) >= SLJIT_EQUAL && (type & 0xff) <= SLJIT_ORDERED_F64); CHECK_ARGUMENT(FUNCTION_CHECK_IS_REG(dst_reg & ~SLJIT_I32_OP)); FUNCTION_CHECK_SRC(src, srcw); #endif @@ -2974,7 +2974,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_x86_emit_cmov(struct sljit_compiler *co if (SLJIT_UNLIKELY(!!compiler->verbose)) { fprintf(compiler->verbose, " x86_cmov%s %s%s, ", !(dst_reg & SLJIT_I32_OP) ? "" : ".i", - JUMP_PREFIX(type), jump_names[type & 0xff]); + jump_names[type & 0xff], JUMP_POSTFIX(type)); sljit_verbose_reg(compiler, dst_reg & ~SLJIT_I32_OP); fprintf(compiler->verbose, ", "); sljit_verbose_param(compiler, src, srcw); -- cgit v1.2.1