Since the destination of reduction is not a vector register group, there
is no need to apply overlap constraint.
Also confirm Clang:
The mir in LLVM has early clobber:
early-clobber %49:vrm2 = PseudoVWADD_VX_M1 $noreg(tied-def 0), killed %17:vr, %48:gpr, %0:gprnox0, 3, 0; example.c:59:24
The mir in LLVM doesn't have early clobber:
%48:vr = PseudoVWREDSUM_VS_M2_E8 $noreg(tied-def 0), %17:vrm2, killed %33:vr, %0:gprnox0, 3, 1; example.c:60:26
And also confirm both:
vwredsum.vs v24, v8, v24 and vwredsum.vs v8, v8, v24 all legal on LLVM.
Align with LLVM and honor RISC-V V spec, remove earlyclobber.
Before this patch:
vwredsum.vs v8,v24,v8
vwredsum.vs v7,v22,v7
vwredsum.vs v6,v20,v6
vwredsum.vs v5,v18,v5
vwredsum.vs v4,v16,v4
vwredsum.vs v3,v14,v3
vwredsum.vs v2,v12,v2
vwredsum.vs v1,v10,v1
vmv1r.v v9,v8
vwredsum.vs v9,v24,v9
vmv1r.v v24,v7
vwredsum.vs v24,v22,v24
vmv1r.v v22,v6
vwredsum.vs v22,v20,v22
vmv1r.v v20,v5
vwredsum.vs v20,v18,v20
vmv1r.v v18,v4
vwredsum.vs v18,v16,v18
vmv1r.v v16,v3
vwredsum.vs v16,v14,v16
vmv1r.v v14,v2
vwredsum.vs v14,v12,v14
vmv1r.v v12,v1
vwredsum.vs v12,v10,v12
After this patch:
vfwredusum.vs v17,v12,v17
vfwredusum.vs v18,v10,v18
vfwredusum.vs v15,v26,v15
vfwredusum.vs v16,v24,v16
vfwredusum.vs v12,v12,v17
vfwredusum.vs v10,v10,v18
vfwredusum.vs v13,v6,v20
vfwredusum.vs v11,v8,v19
vfwredusum.vs v6,v6,v13
vfwredusum.vs v8,v8,v11
vfwredusum.vs v7,v4,v21
vfwredusum.vs v9,v2,v22
vfwredusum.vs v14,v26,v15
vfwredusum.vs v1,v24,v16
vfwredusum.vs v4,v4,v7
vfwredusum.vs v2,v2,v9
Same behavior as LLVM, and honor RISC-V V spec.
PR 112431
gcc/ChangeLog:
* config/riscv/vector.md: Remove earlyclobber from widen reduction.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/base/pr112431-35.c: New test.
* gcc.target/riscv/rvv/base/pr112431-36.c: New test.
---
gcc/config/riscv/vector.md | 8 +-
.../gcc.target/riscv/rvv/base/pr112431-35.c | 107 ++++++++++++++++++
.../gcc.target/riscv/rvv/base/pr112431-36.c | 107 ++++++++++++++++++
3 files changed, 218 insertions(+), 4 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/base/pr112431-35.c
create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/base/pr112431-36.c
@@ -7861,7 +7861,7 @@
;; Integer Widen Reduction Sum (vwredsum[u].vs)
(define_insn "@pred_<reduc_op><mode>"
- [(set (match_operand:<V_EXT_LMUL1> 0 "register_operand" "=&vr,&vr")
+ [(set (match_operand:<V_EXT_LMUL1> 0 "register_operand" "=vr, vr")
(unspec:<V_EXT_LMUL1>
[(unspec:<VM>
[(match_operand:<VM> 1 "vector_mask_operand" "vmWc1,vmWc1")
@@ -7872,7 +7872,7 @@
(reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
(unspec:<V_EXT_LMUL1> [
(match_operand:VI_QHS 3 "register_operand" " vr, vr")
- (match_operand:<V_EXT_LMUL1> 4 "register_operand" " vr0, vr0")
+ (match_operand:<V_EXT_LMUL1> 4 "register_operand" " vr, vr")
] ANY_WREDUC)
(match_operand:<V_EXT_LMUL1> 2 "vector_merge_operand" " vu, 0")] UNSPEC_REDUC))]
"TARGET_VECTOR"
@@ -7928,7 +7928,7 @@
;; Float Widen Reduction Sum (vfwred[ou]sum.vs)
(define_insn "@pred_<reduc_op><mode>"
- [(set (match_operand:<V_EXT_LMUL1> 0 "register_operand" "=&vr, &vr")
+ [(set (match_operand:<V_EXT_LMUL1> 0 "register_operand" "=vr, vr")
(unspec:<V_EXT_LMUL1>
[(unspec:<VM>
[(match_operand:<VM> 1 "vector_mask_operand" "vmWc1,vmWc1")
@@ -7941,7 +7941,7 @@
(reg:SI FRM_REGNUM)] UNSPEC_VPREDICATE)
(unspec:<V_EXT_LMUL1> [
(match_operand:VF_HS 3 "register_operand" " vr, vr")
- (match_operand:<V_EXT_LMUL1> 4 "register_operand" " vr0, vr0")
+ (match_operand:<V_EXT_LMUL1> 4 "register_operand" " vr, vr")
] ANY_FWREDUC_SUM)
(match_operand:<V_EXT_LMUL1> 2 "vector_merge_operand" " vu, 0")] UNSPEC_REDUC))]
"TARGET_VECTOR"
new file mode 100644
@@ -0,0 +1,107 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */
+
+#include "riscv_vector.h"
+
+size_t __attribute__ ((noinline))
+sumation (size_t sum0, size_t sum1, size_t sum2, size_t sum3, size_t sum4,
+ size_t sum5, size_t sum6, size_t sum7,
+ size_t sum0_2, size_t sum1_2, size_t sum2_2, size_t sum3_2, size_t sum4_2,
+ size_t sum5_2, size_t sum6_2, size_t sum7_2)
+{
+ return sum0 + sum1 + sum2 + sum3 + sum4 + sum5 + sum6 + sum7
+ + sum0_2 + sum1_2 + sum2_2 + sum3_2 + sum4_2 + sum5_2 + sum6_2 + sum7_2;
+}
+
+size_t
+foo (char const *buf, size_t len)
+{
+ size_t sum = 0;
+ size_t vl = __riscv_vsetvlmax_e8m8 ();
+ size_t step = vl * 4;
+ const char *it = buf, *end = buf + len;
+ for (; it + step <= end;)
+ {
+ vint8m2_t v0 = __riscv_vle8_v_i8m2 ((void *) it, vl);
+ it += vl;
+ vint8m2_t v1 = __riscv_vle8_v_i8m2 ((void *) it, vl);
+ it += vl;
+ vint8m2_t v2 = __riscv_vle8_v_i8m2 ((void *) it, vl);
+ it += vl;
+ vint8m2_t v3 = __riscv_vle8_v_i8m2 ((void *) it, vl);
+ it += vl;
+ vint8m2_t v4 = __riscv_vle8_v_i8m2 ((void *) it, vl);
+ it += vl;
+ vint8m2_t v5 = __riscv_vle8_v_i8m2 ((void *) it, vl);
+ it += vl;
+ vint8m2_t v6 = __riscv_vle8_v_i8m2 ((void *) it, vl);
+ it += vl;
+ vint8m2_t v7 = __riscv_vle8_v_i8m2 ((void *) it, vl);
+ it += vl;
+
+ vint16m1_t vw0 = __riscv_vle16_v_i16m1 ((void *) it, vl);
+ it += vl;
+ vint16m1_t vw1 = __riscv_vle16_v_i16m1 ((void *) it, vl);
+ it += vl;
+ vint16m1_t vw2 = __riscv_vle16_v_i16m1 ((void *) it, vl);
+ it += vl;
+ vint16m1_t vw3 = __riscv_vle16_v_i16m1 ((void *) it, vl);
+ it += vl;
+ vint16m1_t vw4 = __riscv_vle16_v_i16m1 ((void *) it, vl);
+ it += vl;
+ vint16m1_t vw5 = __riscv_vle16_v_i16m1 ((void *) it, vl);
+ it += vl;
+ vint16m1_t vw6 = __riscv_vle16_v_i16m1 ((void *) it, vl);
+ it += vl;
+ vint16m1_t vw7 = __riscv_vle16_v_i16m1 ((void *) it, vl);
+ it += vl;
+
+ asm volatile("nop" ::: "memory");
+ vint16m1_t vw0_2 = __riscv_vwredsum_vs_i8m2_i16m1 (v0, vw0, vl);
+ vint16m1_t vw1_2 = __riscv_vwredsum_vs_i8m2_i16m1 (v1, vw1, vl);
+ vint16m1_t vw2_2 = __riscv_vwredsum_vs_i8m2_i16m1 (v2, vw2, vl);
+ vint16m1_t vw3_2 = __riscv_vwredsum_vs_i8m2_i16m1 (v3, vw3, vl);
+ vint16m1_t vw4_2 = __riscv_vwredsum_vs_i8m2_i16m1 (v4, vw4, vl);
+ vint16m1_t vw5_2 = __riscv_vwredsum_vs_i8m2_i16m1 (v5, vw5, vl);
+ vint16m1_t vw6_2 = __riscv_vwredsum_vs_i8m2_i16m1 (v6, vw6, vl);
+ vint16m1_t vw7_2 = __riscv_vwredsum_vs_i8m2_i16m1 (v7, vw7, vl);
+
+ vw0 = __riscv_vwredsum_vs_i8m2_i16m1 (v0, vw0_2, vl);
+ vw1 = __riscv_vwredsum_vs_i8m2_i16m1 (v1, vw1_2, vl);
+ vw2 = __riscv_vwredsum_vs_i8m2_i16m1 (v2, vw2_2, vl);
+ vw3 = __riscv_vwredsum_vs_i8m2_i16m1 (v3, vw3_2, vl);
+ vw4 = __riscv_vwredsum_vs_i8m2_i16m1 (v4, vw4_2, vl);
+ vw5 = __riscv_vwredsum_vs_i8m2_i16m1 (v5, vw5_2, vl);
+ vw6 = __riscv_vwredsum_vs_i8m2_i16m1 (v6, vw6_2, vl);
+ vw7 = __riscv_vwredsum_vs_i8m2_i16m1 (v7, vw7_2, vl);
+
+ asm volatile("nop" ::: "memory");
+ size_t sum0 = __riscv_vmv_x_s_i16m1_i16 (vw0);
+ size_t sum1 = __riscv_vmv_x_s_i16m1_i16 (vw1);
+ size_t sum2 = __riscv_vmv_x_s_i16m1_i16 (vw2);
+ size_t sum3 = __riscv_vmv_x_s_i16m1_i16 (vw3);
+ size_t sum4 = __riscv_vmv_x_s_i16m1_i16 (vw4);
+ size_t sum5 = __riscv_vmv_x_s_i16m1_i16 (vw5);
+ size_t sum6 = __riscv_vmv_x_s_i16m1_i16 (vw6);
+ size_t sum7 = __riscv_vmv_x_s_i16m1_i16 (vw7);
+
+ size_t sum0_2 = __riscv_vmv_x_s_i16m1_i16 (vw0_2);
+ size_t sum1_2 = __riscv_vmv_x_s_i16m1_i16 (vw1_2);
+ size_t sum2_2 = __riscv_vmv_x_s_i16m1_i16 (vw2_2);
+ size_t sum3_2 = __riscv_vmv_x_s_i16m1_i16 (vw3_2);
+ size_t sum4_2 = __riscv_vmv_x_s_i16m1_i16 (vw4_2);
+ size_t sum5_2 = __riscv_vmv_x_s_i16m1_i16 (vw5_2);
+ size_t sum6_2 = __riscv_vmv_x_s_i16m1_i16 (vw6_2);
+ size_t sum7_2 = __riscv_vmv_x_s_i16m1_i16 (vw7_2);
+
+ sum += sumation (sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7,
+ sum0_2, sum1_2, sum2_2, sum3_2, sum4_2, sum5_2, sum6_2, sum7_2);
+ }
+ return sum;
+}
+
+/* { dg-final { scan-assembler-not {vmv1r} } } */
+/* { dg-final { scan-assembler-not {vmv2r} } } */
+/* { dg-final { scan-assembler-not {vmv4r} } } */
+/* { dg-final { scan-assembler-not {vmv8r} } } */
+/* { dg-final { scan-assembler-not {csrr} } } */
new file mode 100644
@@ -0,0 +1,107 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */
+
+#include "riscv_vector.h"
+
+size_t __attribute__ ((noinline))
+sumation (size_t sum0, size_t sum1, size_t sum2, size_t sum3, size_t sum4,
+ size_t sum5, size_t sum6, size_t sum7,
+ size_t sum0_2, size_t sum1_2, size_t sum2_2, size_t sum3_2, size_t sum4_2,
+ size_t sum5_2, size_t sum6_2, size_t sum7_2)
+{
+ return sum0 + sum1 + sum2 + sum3 + sum4 + sum5 + sum6 + sum7
+ + sum0_2 + sum1_2 + sum2_2 + sum3_2 + sum4_2 + sum5_2 + sum6_2 + sum7_2;
+}
+
+size_t
+foo (char const *buf, size_t len)
+{
+ size_t sum = 0;
+ size_t vl = __riscv_vsetvlmax_e8m8 ();
+ size_t step = vl * 4;
+ const char *it = buf, *end = buf + len;
+ for (; it + step <= end;)
+ {
+ vfloat32m2_t v0 = __riscv_vle32_v_f32m2 ((void *) it, vl);
+ it += vl;
+ vfloat32m2_t v1 = __riscv_vle32_v_f32m2 ((void *) it, vl);
+ it += vl;
+ vfloat32m2_t v2 = __riscv_vle32_v_f32m2 ((void *) it, vl);
+ it += vl;
+ vfloat32m2_t v3 = __riscv_vle32_v_f32m2 ((void *) it, vl);
+ it += vl;
+ vfloat32m2_t v4 = __riscv_vle32_v_f32m2 ((void *) it, vl);
+ it += vl;
+ vfloat32m2_t v5 = __riscv_vle32_v_f32m2 ((void *) it, vl);
+ it += vl;
+ vfloat32m2_t v6 = __riscv_vle32_v_f32m2 ((void *) it, vl);
+ it += vl;
+ vfloat32m2_t v7 = __riscv_vle32_v_f32m2 ((void *) it, vl);
+ it += vl;
+
+ vfloat64m1_t vw0 = __riscv_vle64_v_f64m1 ((void *) it, vl);
+ it += vl;
+ vfloat64m1_t vw1 = __riscv_vle64_v_f64m1 ((void *) it, vl);
+ it += vl;
+ vfloat64m1_t vw2 = __riscv_vle64_v_f64m1 ((void *) it, vl);
+ it += vl;
+ vfloat64m1_t vw3 = __riscv_vle64_v_f64m1 ((void *) it, vl);
+ it += vl;
+ vfloat64m1_t vw4 = __riscv_vle64_v_f64m1 ((void *) it, vl);
+ it += vl;
+ vfloat64m1_t vw5 = __riscv_vle64_v_f64m1 ((void *) it, vl);
+ it += vl;
+ vfloat64m1_t vw6 = __riscv_vle64_v_f64m1 ((void *) it, vl);
+ it += vl;
+ vfloat64m1_t vw7 = __riscv_vle64_v_f64m1 ((void *) it, vl);
+ it += vl;
+
+ asm volatile("nop" ::: "memory");
+ vfloat64m1_t vw0_2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v0, vw0, vl);
+ vfloat64m1_t vw1_2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v1, vw1, vl);
+ vfloat64m1_t vw2_2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v2, vw2, vl);
+ vfloat64m1_t vw3_2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v3, vw3, vl);
+ vfloat64m1_t vw4_2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v4, vw4, vl);
+ vfloat64m1_t vw5_2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v5, vw5, vl);
+ vfloat64m1_t vw6_2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v6, vw6, vl);
+ vfloat64m1_t vw7_2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v7, vw7, vl);
+
+ vw0 = __riscv_vfwredusum_vs_f32m2_f64m1 (v0, vw0_2, vl);
+ vw1 = __riscv_vfwredusum_vs_f32m2_f64m1 (v1, vw1_2, vl);
+ vw2 = __riscv_vfwredusum_vs_f32m2_f64m1 (v2, vw2_2, vl);
+ vw3 = __riscv_vfwredusum_vs_f32m2_f64m1 (v3, vw3_2, vl);
+ vw4 = __riscv_vfwredusum_vs_f32m2_f64m1 (v4, vw4_2, vl);
+ vw5 = __riscv_vfwredusum_vs_f32m2_f64m1 (v5, vw5_2, vl);
+ vw6 = __riscv_vfwredusum_vs_f32m2_f64m1 (v6, vw6_2, vl);
+ vw7 = __riscv_vfwredusum_vs_f32m2_f64m1 (v7, vw7_2, vl);
+
+ asm volatile("nop" ::: "memory");
+ size_t sum0 = __riscv_vfmv_f_s_f64m1_f64 (vw0);
+ size_t sum1 = __riscv_vfmv_f_s_f64m1_f64 (vw1);
+ size_t sum2 = __riscv_vfmv_f_s_f64m1_f64 (vw2);
+ size_t sum3 = __riscv_vfmv_f_s_f64m1_f64 (vw3);
+ size_t sum4 = __riscv_vfmv_f_s_f64m1_f64 (vw4);
+ size_t sum5 = __riscv_vfmv_f_s_f64m1_f64 (vw5);
+ size_t sum6 = __riscv_vfmv_f_s_f64m1_f64 (vw6);
+ size_t sum7 = __riscv_vfmv_f_s_f64m1_f64 (vw7);
+
+ size_t sum0_2 = __riscv_vfmv_f_s_f64m1_f64 (vw0_2);
+ size_t sum1_2 = __riscv_vfmv_f_s_f64m1_f64 (vw1_2);
+ size_t sum2_2 = __riscv_vfmv_f_s_f64m1_f64 (vw2_2);
+ size_t sum3_2 = __riscv_vfmv_f_s_f64m1_f64 (vw3_2);
+ size_t sum4_2 = __riscv_vfmv_f_s_f64m1_f64 (vw4_2);
+ size_t sum5_2 = __riscv_vfmv_f_s_f64m1_f64 (vw5_2);
+ size_t sum6_2 = __riscv_vfmv_f_s_f64m1_f64 (vw6_2);
+ size_t sum7_2 = __riscv_vfmv_f_s_f64m1_f64 (vw7_2);
+
+ sum += sumation (sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7,
+ sum0_2, sum1_2, sum2_2, sum3_2, sum4_2, sum5_2, sum6_2, sum7_2);
+ }
+ return sum;
+}
+
+/* { dg-final { scan-assembler-not {vmv1r} } } */
+/* { dg-final { scan-assembler-not {vmv2r} } } */
+/* { dg-final { scan-assembler-not {vmv4r} } } */
+/* { dg-final { scan-assembler-not {vmv8r} } } */
+/* { dg-final { scan-assembler-not {csrr} } } */