Skip to content

Commit c87e1be

Browse files
author
Srinivas Vamsi Parasa
committed
8349582: APX NDD code generation for OpenJDK
Reviewed-by: epeter, jbhateja, sviswanathan
1 parent 7d1fe0e commit c87e1be

File tree

5 files changed

+3572
-1606
lines changed

5 files changed

+3572
-1606
lines changed

‎src/hotspot/cpu/x86/assembler_x86.cpp

+25-10
Original file line numberDiff line numberDiff line change
@@ -332,6 +332,21 @@ void Assembler::emit_arith(int op1, int op2, Register dst, int32_t imm32) {
332332
}
333333
}
334334

335+
void Assembler::emit_arith_ndd(int op1, int op2, Register dst, int32_t imm32) {
336+
assert(isByte(op1) && isByte(op2), "wrong opcode");
337+
assert(op1 == 0x81, "Unexpected opcode");
338+
// This code cache friendly optimization saves 3 bytes per encoding, which offsets the EVEX encoding penalty.
339+
if (is8bit(imm32)) {
340+
emit_int24(op1 | 0x02, // set sign bit
341+
op2 | encode(dst),
342+
imm32 & 0xFF);
343+
}
344+
else {
345+
emit_int16(op1, (op2 | encode(dst)));
346+
emit_int32(imm32);
347+
}
348+
}
349+
335350
// Force generation of a 4 byte immediate value even if it fits into 8bit
336351
void Assembler::emit_arith_imm32(int op1, int op2, Register dst, int32_t imm32) {
337352
assert(isByte(op1) && isByte(op2), "wrong opcode");
@@ -1461,7 +1476,7 @@ void Assembler::addl(Register dst, int32_t imm32) {
14611476
void Assembler::eaddl(Register dst, Register src, int32_t imm32, bool no_flags) {
14621477
InstructionAttr attributes(AVX_128bit, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
14631478
(void) evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
1464-
emit_arith(0x81, 0xC0, src, imm32);
1479+
emit_arith_ndd(0x81, 0xC0, src, imm32);
14651480
}
14661481

14671482
void Assembler::addl(Register dst, Address src) {
@@ -1695,7 +1710,7 @@ void Assembler::andl(Register dst, int32_t imm32) {
16951710
void Assembler::eandl(Register dst, Register src, int32_t imm32, bool no_flags) {
16961711
InstructionAttr attributes(AVX_128bit, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
16971712
(void) evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
1698-
emit_arith(0x81, 0xE0, src, imm32);
1713+
emit_arith_ndd(0x81, 0xE0, src, imm32);
16991714
}
17001715

17011716
void Assembler::andl(Address dst, Register src) {
@@ -4532,7 +4547,7 @@ void Assembler::orl(Register dst, int32_t imm32) {
45324547
void Assembler::eorl(Register dst, Register src, int32_t imm32, bool no_flags) {
45334548
InstructionAttr attributes(AVX_128bit, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
45344549
evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
4535-
emit_arith(0x81, 0xC8, src, imm32);
4550+
emit_arith_ndd(0x81, 0xC8, src, imm32);
45364551
}
45374552

45384553
void Assembler::orl(Register dst, Address src) {
@@ -7171,7 +7186,7 @@ void Assembler::subl(Register dst, int32_t imm32) {
71717186
void Assembler::esubl(Register dst, Register src, int32_t imm32, bool no_flags) {
71727187
InstructionAttr attributes(AVX_128bit, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
71737188
(void) evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
7174-
emit_arith(0x81, 0xE8, src, imm32);
7189+
emit_arith_ndd(0x81, 0xE8, src, imm32);
71757190
}
71767191

71777192
// Force generation of a 4 byte immediate value even if it fits into 8bit
@@ -7512,7 +7527,7 @@ void Assembler::xorl(Register dst, int32_t imm32) {
75127527
void Assembler::exorl(Register dst, Register src, int32_t imm32, bool no_flags) {
75137528
InstructionAttr attributes(AVX_128bit, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
75147529
evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
7515-
emit_arith(0x81, 0xF0, src, imm32);
7530+
emit_arith_ndd(0x81, 0xF0, src, imm32);
75167531
}
75177532

75187533
void Assembler::xorl(Register dst, Address src) {
@@ -15158,7 +15173,7 @@ void Assembler::addq(Register dst, int32_t imm32) {
1515815173
void Assembler::eaddq(Register dst, Register src, int32_t imm32, bool no_flags) {
1515915174
InstructionAttr attributes(AVX_128bit, /* vex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
1516015175
(void) evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
15161-
emit_arith(0x81, 0xC0, src, imm32);
15176+
emit_arith_ndd(0x81, 0xC0, src, imm32);
1516215177
}
1516315178

1516415179
void Assembler::addq(Register dst, Address src) {
@@ -15255,7 +15270,7 @@ void Assembler::andq(Register dst, int32_t imm32) {
1525515270
void Assembler::eandq(Register dst, Register src, int32_t imm32, bool no_flags) {
1525615271
InstructionAttr attributes(AVX_128bit, /* vex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
1525715272
evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
15258-
emit_arith(0x81, 0xE0, src, imm32);
15273+
emit_arith_ndd(0x81, 0xE0, src, imm32);
1525915274
}
1526015275

1526115276
void Assembler::andq(Register dst, Address src) {
@@ -16142,7 +16157,7 @@ void Assembler::orq(Register dst, int32_t imm32) {
1614216157
void Assembler::eorq(Register dst, Register src, int32_t imm32, bool no_flags) {
1614316158
InstructionAttr attributes(AVX_128bit, /* vex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
1614416159
evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
16145-
emit_arith(0x81, 0xC8, src, imm32);
16160+
emit_arith_ndd(0x81, 0xC8, src, imm32);
1614616161
}
1614716162

1614816163
void Assembler::orq_imm32(Register dst, int32_t imm32) {
@@ -16830,7 +16845,7 @@ void Assembler::subq(Register dst, int32_t imm32) {
1683016845
void Assembler::esubq(Register dst, Register src, int32_t imm32, bool no_flags) {
1683116846
InstructionAttr attributes(AVX_128bit, /* vex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
1683216847
(void) evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
16833-
emit_arith(0x81, 0xE8, src, imm32);
16848+
emit_arith_ndd(0x81, 0xE8, src, imm32);
1683416849
}
1683516850

1683616851
// Force generation of a 4 byte immediate value even if it fits into 8bit
@@ -16961,7 +16976,7 @@ void Assembler::xorq(Register dst, int32_t imm32) {
1696116976
void Assembler::exorq(Register dst, Register src, int32_t imm32, bool no_flags) {
1696216977
InstructionAttr attributes(AVX_128bit, /* vex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
1696316978
evex_prefix_and_encode_ndd(0, dst->encoding(), src->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F_3C, &attributes, no_flags);
16964-
emit_arith(0x81, 0xF0, src, imm32);
16979+
emit_arith_ndd(0x81, 0xF0, src, imm32);
1696516980
}
1696616981

1696716982
void Assembler::xorq(Address dst, int32_t imm32) {

‎src/hotspot/cpu/x86/assembler_x86.hpp

+1
Original file line numberDiff line numberDiff line change
@@ -823,6 +823,7 @@ class Assembler : public AbstractAssembler {
823823
void emit_arith_b(int op1, int op2, Register dst, int imm8);
824824

825825
void emit_arith(int op1, int op2, Register dst, int32_t imm32);
826+
void emit_arith_ndd(int op1, int op2, Register dst, int32_t imm32);
826827
// Force generation of a 4 byte immediate value even if it fits into 8bit
827828
void emit_arith_imm32(int op1, int op2, Register dst, int32_t imm32);
828829
void emit_arith(int op1, int op2, Register dst, Register src);

0 commit comments

Comments
 (0)