1berry/openjdk
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

8356593: RISC-V: Small improvement to array fill stub

Browse Source 
Reviewed-by: fyang
This commit is contained in:
Anjian-Wen 2025-05-12 03:01:46 +00:00 committed by Fei Yang
parent 74f047b84d
commit d7cb933b89
1 changed files with 46 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

79
src/hotspot/cpu/riscv/stubGenerator_riscv.cpp
View File

@ -2029,44 +2029,40 @@ class StubGenerator: public StubCodeGenerator {
__ enter(); __ enter();
Label L_fill_elements, L_exit1; Label L_fill_elements;
int shift = -1; int shift = -1;
switch (t) { switch (t) {
case T_BYTE: case T_BYTE:
shift = 0; shift = 0;
// Short arrays (< 8 bytes) fill by element
__ mv(tmp_reg, 8 >> shift);
__ bltu(count, tmp_reg, L_fill_elements);
// Zero extend value // Zero extend value
// 8 bit -> 16 bit // 8 bit -> 16 bit
__ zext(value, value, 8); __ zext(value, value, 8);
__ mv(tmp_reg, value); __ slli(tmp_reg, value, 8);
__ slli(tmp_reg, tmp_reg, 8);
__ orr(value, value, tmp_reg); __ orr(value, value, tmp_reg);
// 16 bit -> 32 bit // 16 bit -> 32 bit
__ mv(tmp_reg, value); __ slli(tmp_reg, value, 16);
__ slli(tmp_reg, tmp_reg, 16);
__ orr(value, value, tmp_reg); __ orr(value, value, tmp_reg);
__ mv(tmp_reg, 8 >> shift); // Short arrays (< 8 bytes) fill by element
__ bltu(count, tmp_reg, L_fill_elements);
break; break;
case T_SHORT: case T_SHORT:
shift = 1; shift = 1;
// Zero extend value
// 16 bit -> 32 bit
__ zext(value, value, 16);
__ mv(tmp_reg, value);
__ slli(tmp_reg, tmp_reg, 16);
__ orr(value, value, tmp_reg);
// Short arrays (< 8 bytes) fill by element // Short arrays (< 8 bytes) fill by element
__ mv(tmp_reg, 8 >> shift); __ mv(tmp_reg, 8 >> shift);
__ bltu(count, tmp_reg, L_fill_elements); __ bltu(count, tmp_reg, L_fill_elements);
// Zero extend value
// 16 bit -> 32 bit
__ zext(value, value, 16);
__ slli(tmp_reg, value, 16);
__ orr(value, value, tmp_reg);
break; break;
case T_INT: case T_INT:
shift = 2; shift = 2;
// Short arrays (< 8 bytes) fill by element // Short arrays (< 8 bytes) fill by element
__ mv(tmp_reg, 8 >> shift); __ mv(tmp_reg, 8 >> shift);
__ bltu(count, tmp_reg, L_fill_elements); __ bltu(count, tmp_reg, L_fill_elements);
@ -2125,20 +2121,8 @@ class StubGenerator: public StubCodeGenerator {
__ fill_words(to, cnt_words, value); __ fill_words(to, cnt_words, value);
} }
// Remaining count is less than 8 bytes. Fill it by a single store. // Remaining count is less than 8 bytes and address is heapword aligned.
// Note that the total length is no less than 8 bytes. Label L_fill_2, L_fill_4, L_exit1;
if (!AvoidUnalignedAccesses && (t == T_BYTE || t == T_SHORT)) {
__ beqz(count, L_exit1);
__ shadd(to, count, to, tmp_reg, shift); // points to the end
__ sd(value, Address(to, -8)); // overwrite some elements
__ bind(L_exit1);
__ leave();
__ ret();
}
// Handle copies less than 8 bytes.
Label L_fill_2, L_fill_4, L_exit2;
__ bind(L_fill_elements);
switch (t) { switch (t) {
case T_BYTE: case T_BYTE:
__ test_bit(t0, count, 0); __ test_bit(t0, count, 0);
@ -2152,7 +2136,7 @@ class StubGenerator: public StubCodeGenerator {
__ addi(to, to, 2); __ addi(to, to, 2);
__ bind(L_fill_4); __ bind(L_fill_4);
__ test_bit(t0, count, 2); __ test_bit(t0, count, 2);
__ beqz(t0, L_exit2); __ beqz(t0, L_exit1);
__ sw(value, Address(to, 0)); __ sw(value, Address(to, 0));
break; break;
case T_SHORT: case T_SHORT:
@ -2162,11 +2146,39 @@ class StubGenerator: public StubCodeGenerator {
__ addi(to, to, 2); __ addi(to, to, 2);
__ bind(L_fill_4); __ bind(L_fill_4);
__ test_bit(t0, count, 1); __ test_bit(t0, count, 1);
__ beqz(t0, L_exit2); __ beqz(t0, L_exit1);
__ sw(value, Address(to, 0)); __ sw(value, Address(to, 0));
break; break;
case T_INT: case T_INT:
__ beqz(count, L_exit2); __ beqz(count, L_exit1);
__ sw(value, Address(to, 0));
break;
default: ShouldNotReachHere();
}
__ bind(L_exit1);
__ leave();
__ ret();
// Handle copies less than 8 bytes.
Label L_loop1, L_loop2, L_exit2;
__ bind(L_fill_elements);
__ beqz(count, L_exit2);
switch (t) {
case T_BYTE:
__ bind(L_loop1);
__ sb(value, Address(to, 0));
__ addi(to, to, 1);
__ subiw(count, count, 1);
__ bnez(count, L_loop1);
break;
case T_SHORT:
__ bind(L_loop2);
__ sh(value, Address(to, 0));
__ addi(to, to, 2);
__ subiw(count, count, 2 >> shift);
__ bnez(count, L_loop2);
break;
case T_INT:
__ sw(value, Address(to, 0)); __ sw(value, Address(to, 0));
break; break;
default: ShouldNotReachHere(); default: ShouldNotReachHere();
@ -2174,6 +2186,7 @@ class StubGenerator: public StubCodeGenerator {
__ bind(L_exit2); __ bind(L_exit2);
__ leave(); __ leave();
__ ret(); __ ret();
return start; return start;
} }