[3/3] RISC-V: Add memset-zero expansion to cbo.zero
Checks
Context |
Check |
Description |
rivoscibot/toolchain-ci-rivos-lint |
success
|
Lint passed
|
rivoscibot/toolchain-ci-rivos-apply-patch |
success
|
Patch applied
|
rivoscibot/toolchain-ci-rivos-build--linux-rv64gc_zba_zbb_zbc_zbs-lp64d-non-multilib |
fail
|
Build failed
|
rivoscibot/toolchain-ci-rivos-build--newlib-rv64gc-lp64d-multilib |
fail
|
Build failed
|
rivoscibot/toolchain-ci-rivos-build--linux-rv32gc_zba_zbb_zbc_zbs-ilp32d-non-multilib |
fail
|
Build failed
|
linaro-tcwg-bot/tcwg_gcc_build--master-arm |
success
|
Testing passed
|
linaro-tcwg-bot/tcwg_gcc_build--master-aarch64 |
success
|
Testing passed
|
rivoscibot/toolchain-ci-rivos-build--linux-rv64gcv-lp64d-multilib |
fail
|
Build failed
|
rivoscibot/toolchain-ci-rivos-build--newlib-rv64gcv-lp64d-multilib |
fail
|
Build failed
|
rivoscibot/toolchain-ci-rivos-test |
fail
|
Testing failed
|
Commit Message
The Zicboz extension offers the cbo.zero instruction, which can be used
to clean a memory region corresponding to a cache block.
The Zic64b extension defines the cache block size to 64 byte.
If both extensions are available, it is possible to use cbo.zero
to clear memory, if the alignment and size constraints are met.
This patch implements this.
gcc/ChangeLog:
* config/riscv/riscv-protos.h (riscv_expand_block_clear): New prototype.
* config/riscv/riscv-string.cc (riscv_expand_block_clear_zicboz_zic64b):
New function to expand a block-clear with cbo.zero.
(riscv_expand_block_clear): New RISC-V block-clear expansion function.
* config/riscv/riscv.md (setmem<mode>): New setmem expansion.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/cmo-zicboz-zic64-1.c: New test.
Signed-off-by: Christoph Müllner <christoph.muellner@vrull.eu>
---
gcc/config/riscv/riscv-protos.h | 1 +
gcc/config/riscv/riscv-string.cc | 59 +++++++++++++++++++
gcc/config/riscv/riscv.md | 24 ++++++++
.../gcc.target/riscv/cmo-zicboz-zic64-1.c | 43 ++++++++++++++
4 files changed, 127 insertions(+)
create mode 100644 gcc/testsuite/gcc.target/riscv/cmo-zicboz-zic64-1.c
Comments
On 5/7/24 11:38 PM, Christoph Müllner wrote:
> The Zicboz extension offers the cbo.zero instruction, which can be used
> to clean a memory region corresponding to a cache block.
> The Zic64b extension defines the cache block size to 64 byte.
> If both extensions are available, it is possible to use cbo.zero
> to clear memory, if the alignment and size constraints are met.
> This patch implements this.
>
> gcc/ChangeLog:
>
> * config/riscv/riscv-protos.h (riscv_expand_block_clear): New prototype.
> * config/riscv/riscv-string.cc (riscv_expand_block_clear_zicboz_zic64b):
> New function to expand a block-clear with cbo.zero.
> (riscv_expand_block_clear): New RISC-V block-clear expansion function.
> * config/riscv/riscv.md (setmem<mode>): New setmem expansion.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/riscv/cmo-zicboz-zic64-1.c: New test.
Depending on the underlying uarch details cbo.zero may not be nearly as
useful as it might first appear. There can be multiple uarch details
that come into play. We've done a fair amount of measurement internally
in this space and while cbo.zero is a win, it's not a huge win. Point
being we may nee to come back and make this part of the tuning structure
so uarchs can adjust.
--
I know in the cbo memset implementation VRULL provided to Ventana you
used the trick of allowing overlapping stores to avoid the alignment
requirements. ie we issue a series of "sd" instrutions to ensure we
cross the alignment barrier, then a series of cbo.zero instructions for
the cache lines (possibly overlapping the locations stored by those "sd"
instructions, then handled residuals which may overlap the last cbo.zero
instructions.
I don't think you necessarily have to do that for this patch, but I
suspect that a similar approach would make this apply much more often in
practice.
So, OK for the trunk and consider the unaligned cases as potential
follow-up enhancements.
THanks
Jeff
@@ -189,6 +189,7 @@ rtl_opt_pass * make_pass_vsetvl (gcc::context *ctxt);
/* Routines implemented in riscv-string.c. */
extern bool riscv_expand_block_move (rtx, rtx, rtx);
+extern bool riscv_expand_block_clear (rtx, rtx);
/* Information about one CPU we know about. */
struct riscv_cpu_info {
@@ -787,6 +787,65 @@ riscv_expand_block_move (rtx dest, rtx src, rtx length)
return false;
}
+/* Expand a block-clear instruction via cbo.zero instructions. */
+
+static bool
+riscv_expand_block_clear_zicboz_zic64b (rtx dest, rtx length)
+{
+ unsigned HOST_WIDE_INT hwi_length;
+ unsigned HOST_WIDE_INT align;
+ const unsigned HOST_WIDE_INT cbo_bytes = 64;
+
+ gcc_assert (TARGET_ZICBOZ && TARGET_ZIC64B);
+
+ if (!CONST_INT_P (length))
+ return false;
+
+ hwi_length = UINTVAL (length);
+ if (hwi_length < cbo_bytes)
+ return false;
+
+ align = MEM_ALIGN (dest) / BITS_PER_UNIT;
+ if (align < cbo_bytes)
+ return false;
+
+ /* We don't emit loops. Instead apply move-bytes limitation. */
+ unsigned HOST_WIDE_INT max_bytes = RISCV_MAX_MOVE_BYTES_STRAIGHT /
+ UNITS_PER_WORD * cbo_bytes;
+ if (hwi_length > max_bytes)
+ return false;
+
+ unsigned HOST_WIDE_INT offset = 0;
+ while (offset + cbo_bytes <= hwi_length)
+ {
+ rtx mem = adjust_address (dest, BLKmode, offset);
+ rtx addr = force_reg (Pmode, XEXP (mem, 0));
+ emit_insn (gen_riscv_zero_di (addr));
+ offset += cbo_bytes;
+ }
+
+ if (offset < hwi_length)
+ {
+ rtx mem = adjust_address (dest, BLKmode, offset);
+ clear_by_pieces (mem, hwi_length - offset, align);
+ }
+
+ return true;
+}
+
+bool
+riscv_expand_block_clear (rtx dest, rtx length)
+{
+ /* Only use setmem-zero expansion for Zicboz + Zic64b. */
+ if (!TARGET_ZICBOZ || !TARGET_ZIC64B)
+ return false;
+
+ if (optimize_function_for_size_p (cfun))
+ return false;
+
+ return riscv_expand_block_clear_zicboz_zic64b (dest, length);
+}
+
/* --- Vector expanders --- */
namespace riscv_vector {
@@ -2598,6 +2598,30 @@ (define_expand "cpymem<mode>"
FAIL;
})
+;; Fill memory with constant byte.
+;; Argument 0 is the destination
+;; Argument 1 is the constant byte
+;; Argument 2 is the length
+;; Argument 3 is the alignment
+
+(define_expand "setmem<mode>"
+ [(parallel [(set (match_operand:BLK 0 "memory_operand")
+ (match_operand:QI 2 "const_int_operand"))
+ (use (match_operand:P 1 ""))
+ (use (match_operand:SI 3 "const_int_operand"))])]
+ ""
+ {
+ /* If value to set is not zero, use the library routine. */
+ if (operands[2] != const0_rtx)
+ FAIL;
+
+ if (riscv_expand_block_clear (operands[0], operands[1]))
+ DONE;
+ else
+ FAIL;
+})
+
+
;; Expand in-line code to clear the instruction cache between operand[0] and
;; operand[1].
(define_expand "clear_cache"
new file mode 100644
@@ -0,0 +1,43 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gc_zic64b_zicboz" { target { rv64 } } } */
+/* { dg-options "-march=rv32gc_zic64b_zicboz" { target { rv32 } } } */
+/* { dg-skip-if "" { *-*-* } {"-O0" "-Os" "-Og" "-Oz" "-flto" } } */
+/* { dg-final { check-function-bodies "**" "" } } */
+/* { dg-allow-blank-lines-in-output 1 } */
+
+/*
+**clear_buf_123:
+** ...
+** cbo\.zero\t0\(a[0-9]+\)
+** sd\tzero,64\(a[0-9]+\)
+** sd\tzero,72\(a[0-9]+\)
+** sd\tzero,80\(a[0-9]+\)
+** sd\tzero,88\(a[0-9]+\)
+** sd\tzero,96\(a[0-9]+\)
+** sd\tzero,104\(a[0-9]+\)
+** sd\tzero,112\(a[0-9]+\)
+** sh\tzero,120\(a[0-9]+\)
+** sb\tzero,122\(a[0-9]+\)
+** ...
+*/
+int
+clear_buf_123 (void *p)
+{
+ p = __builtin_assume_aligned(p, 64);
+ __builtin_memset (p, 0, 123);
+}
+
+/*
+**clear_buf_128:
+** ...
+** cbo\.zero\t0\(a[0-9]+\)
+** addi\ta[0-9]+,a[0-9]+,64
+** cbo\.zero\t0\(a[0-9]+\)
+** ...
+*/
+int
+clear_buf_128 (void *p)
+{
+ p = __builtin_assume_aligned(p, 64);
+ __builtin_memset (p, 0, 128);
+}