UNSPEC_DMF_PMDMXVBF16GERX2PN
UNSPEC_DMF_PMDMXVBF16GERX2NP
UNSPEC_DMF_PMDMXVBF16GERX2NN
+ UNSPEC_DMF_DMXVF16GERX2
+ UNSPEC_DMF_DMXVF16GERX2PP
+ UNSPEC_DMF_DMXVF16GERX2PN
+ UNSPEC_DMF_DMXVF16GERX2NP
+ UNSPEC_DMF_DMXVF16GERX2NN
+ UNSPEC_DMF_PMDMXVF16GERX2
+ UNSPEC_DMF_PMDMXVF16GERX2PP
+ UNSPEC_DMF_PMDMXVF16GERX2PN
+ UNSPEC_DMF_PMDMXVF16GERX2NP
+ UNSPEC_DMF_PMDMXVF16GERX2NN
])
(define_c_enum "unspecv"
; DMF instructions with 1 vector pair and 1 vector arguments
(define_int_iterator DMF_PV [UNSPEC_DMF_DMXVI8GERX4
- UNSPEC_DMF_DMXVBF16GERX2])
+ UNSPEC_DMF_DMXVBF16GERX2
+ UNSPEC_DMF_DMXVF16GERX2])
;; MMA instructions with 1 accumulator, 1 vector pair and 1 vector arguments
(define_int_iterator MMA_APV [UNSPEC_MMA_XVF64GERPP
UNSPEC_DMF_DMXVBF16GERX2PP
UNSPEC_DMF_DMXVBF16GERX2PN
UNSPEC_DMF_DMXVBF16GERX2NP
- UNSPEC_DMF_DMXVBF16GERX2NN])
+ UNSPEC_DMF_DMXVBF16GERX2NN
+ UNSPEC_DMF_DMXVF16GERX2PP
+ UNSPEC_DMF_DMXVF16GERX2PN
+ UNSPEC_DMF_DMXVF16GERX2NP
+ UNSPEC_DMF_DMXVF16GERX2NN])
;; MMA instructions with 2 vector, 2 4-bit and 1 8-bit arguments
(define_int_iterator MMA_VVI4I4I8 [UNSPEC_MMA_PMXVI4GER8])
;; DMF instructions with 1 vector pair, 1 vector, 1 8-bit, 1 4-bit
;; and 1 2-bit arguments
-(define_int_iterator DMF_PVI8I4I2 [UNSPEC_DMF_PMDMXVBF16GERX2])
+(define_int_iterator DMF_PVI8I4I2 [UNSPEC_DMF_PMDMXVBF16GERX2
+ UNSPEC_DMF_PMDMXVF16GERX2])
;; DMF instructions with 1dmr, 1 vector pair, 1 vector, 1 8-bit,
;; 1 4-bit and 1 2-bit arguments
(define_int_iterator DMF_DPVI8I4I2 [UNSPEC_DMF_PMDMXVBF16GERX2PP
UNSPEC_DMF_PMDMXVBF16GERX2PN
UNSPEC_DMF_PMDMXVBF16GERX2NP
- UNSPEC_DMF_PMDMXVBF16GERX2NN])
+ UNSPEC_DMF_PMDMXVBF16GERX2NN
+ UNSPEC_DMF_PMDMXVF16GERX2PP
+ UNSPEC_DMF_PMDMXVF16GERX2PN
+ UNSPEC_DMF_PMDMXVF16GERX2NP
+ UNSPEC_DMF_PMDMXVF16GERX2NN])
(define_int_attr acc [(UNSPEC_MMA_XXMFACC "xxmfacc")
(UNSPEC_MMA_XXMTACC "xxmtacc")])
(define_int_attr pv [(UNSPEC_MMA_XVF64GER "xvf64ger")
(UNSPEC_DMF_DMXVI8GERX4 "dmxvi8gerx4")
- (UNSPEC_DMF_DMXVBF16GERX2 "dmxvbf16gerx2")])
+ (UNSPEC_DMF_DMXVBF16GERX2 "dmxvbf16gerx2")
+ (UNSPEC_DMF_DMXVF16GERX2 "dmxvf16gerx2")])
(define_int_attr apv [(UNSPEC_MMA_XVF64GERPP "xvf64gerpp")
(UNSPEC_MMA_XVF64GERPN "xvf64gerpn")
(UNSPEC_DMF_DMXVBF16GERX2PP "dmxvbf16gerx2pp")
(UNSPEC_DMF_DMXVBF16GERX2PN "dmxvbf16gerx2pn")
(UNSPEC_DMF_DMXVBF16GERX2NP "dmxvbf16gerx2np")
- (UNSPEC_DMF_DMXVBF16GERX2NN "dmxvbf16gerx2nn")])
+ (UNSPEC_DMF_DMXVBF16GERX2NN "dmxvbf16gerx2nn")
+ (UNSPEC_DMF_DMXVF16GERX2PP "dmxvf16gerx2pp")
+ (UNSPEC_DMF_DMXVF16GERX2PN "dmxvf16gerx2pn")
+ (UNSPEC_DMF_DMXVF16GERX2NP "dmxvf16gerx2np")
+ (UNSPEC_DMF_DMXVF16GERX2NN "dmxvf16gerx2nn")])
(define_int_attr vvi4i4i8 [(UNSPEC_MMA_PMXVI4GER8 "pmxvi4ger8")])
(define_int_attr dpvi8i4i4 [(UNSPEC_DMF_PMDMXVI8GERX4PP "pmdmxvi8gerx4pp")
(UNSPEC_DMF_PMDMXVI8GERX4SPP "pmdmxvi8gerx4spp")])
-(define_int_attr pvi8i4i2 [(UNSPEC_DMF_PMDMXVBF16GERX2 "pmdmxvbf16gerx2")])
+(define_int_attr pvi8i4i2 [(UNSPEC_DMF_PMDMXVBF16GERX2 "pmdmxvbf16gerx2")
+ (UNSPEC_DMF_PMDMXVF16GERX2 "pmdmxvf16gerx2")])
(define_int_attr dpvi8i4i2 [(UNSPEC_DMF_PMDMXVBF16GERX2PP "pmdmxvbf16gerx2pp")
(UNSPEC_DMF_PMDMXVBF16GERX2PN "pmdmxvbf16gerx2pn")
(UNSPEC_DMF_PMDMXVBF16GERX2NP "pmdmxvbf16gerx2np")
- (UNSPEC_DMF_PMDMXVBF16GERX2NN "pmdmxvbf16gerx2nn")])
+ (UNSPEC_DMF_PMDMXVBF16GERX2NN "pmdmxvbf16gerx2nn")
+ (UNSPEC_DMF_PMDMXVF16GERX2PP "pmdmxvf16gerx2pp")
+ (UNSPEC_DMF_PMDMXVF16GERX2PN "pmdmxvf16gerx2pn")
+ (UNSPEC_DMF_PMDMXVF16GERX2NP "pmdmxvf16gerx2np")
+ (UNSPEC_DMF_PMDMXVF16GERX2NN "pmdmxvf16gerx2nn")])
;; Vector pair support. OOmode can only live in VSRs.
(define_expand "movoo"
[(set_attr "type" "dmf")])
(define_insn "dmf_<pv>"
- [(set (match_operand:TDO 0 "accumulator_operand" "=wD")
+ [(set (match_operand:TDO 0 "dmr_register_operand" "=wD")
(unspec:TDO [(match_operand:OO 1 "vsx_register_operand" "wa")
(match_operand:V16QI 2 "vsx_register_operand" "wa")]
DMF_PV))]
dmr1024 __builtin_mma_pmdmxvbf16gerx2nn_internal (dmr1024, v256, vuc, const int<8>, \
const int<4>, const int<2>);
PMDMXVBF16GERX2NN_INTERNAL dmf_pmdmxvbf16gerx2nn {dm,pair}
+
+ void __builtin_mma_dmxvf16gerx2 (dmr1024 *, v256, vuc);
+ DMXVF16GERX2 nothing {dm,dmint}
+
+ dmr1024 __builtin_mma_dmxvf16gerx2_internal (v256, vuc);
+ DMXVF16GERX2_INTERNAL dmf_dmxvf16gerx2 {dm}
+
+ void __builtin_mma_dmxvf16gerx2pp (dmr1024 *, v256, vuc);
+ DMXVF16GERX2PP nothing {dm,dmint,dmr}
+
+ dmr1024 __builtin_mma_dmxvf16gerx2pp_internal (dmr1024, v256, vuc);
+ DMXVF16GERX2PP_INTERNAL dmf_dmxvf16gerx2pp {dm}
+
+ void __builtin_mma_dmxvf16gerx2pn (dmr1024 *, v256, vuc);
+ DMXVF16GERX2PN nothing {dm,dmint,dmr}
+
+ dmr1024 __builtin_mma_dmxvf16gerx2pn_internal (dmr1024, v256, vuc);
+ DMXVF16GERX2PN_INTERNAL dmf_dmxvf16gerx2pn {dm}
+
+ void __builtin_mma_dmxvf16gerx2np (dmr1024 *, v256, vuc);
+ DMXVF16GERX2NP nothing {dm,dmint,dmr}
+
+ dmr1024 __builtin_mma_dmxvf16gerx2np_internal (dmr1024, v256, vuc);
+ DMXVF16GERX2NP_INTERNAL dmf_dmxvf16gerx2np {dm}
+
+ void __builtin_mma_dmxvf16gerx2nn (dmr1024 *, v256, vuc);
+ DMXVF16GERX2NN nothing {dm,dmint,dmr}
+
+ dmr1024 __builtin_mma_dmxvf16gerx2nn_internal (dmr1024, v256, vuc);
+ DMXVF16GERX2NN_INTERNAL dmf_dmxvf16gerx2nn {dm}
+
+ void __builtin_mma_pmdmxvf16gerx2 (dmr1024 *, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2 nothing {dm,pair,dmint}
+
+ dmr1024 __builtin_mma_pmdmxvf16gerx2_internal (v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2_INTERNAL dmf_pmdmxvf16gerx2 {dm,pair}
+
+ void __builtin_mma_pmdmxvf16gerx2pp (dmr1024 *, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2PP nothing {dm,pair,dmint,dmr}
+
+ dmr1024 __builtin_mma_pmdmxvf16gerx2pp_internal (dmr1024, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2PP_INTERNAL dmf_pmdmxvf16gerx2pp {dm,pair}
+
+ void __builtin_mma_pmdmxvf16gerx2pn (dmr1024 *, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2PN nothing {dm,pair,dmint,dmr}
+
+ dmr1024 __builtin_mma_pmdmxvf16gerx2pn_internal (dmr1024, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2PN_INTERNAL dmf_pmdmxvf16gerx2pn {dm,pair}
+
+ void __builtin_mma_pmdmxvf16gerx2np (dmr1024 *, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2NP nothing {dm,pair,dmint,dmr}
+
+ dmr1024 __builtin_mma_pmdmxvf16gerx2np_internal (dmr1024, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2NP_INTERNAL dmf_pmdmxvf16gerx2np {dm,pair}
+
+ void __builtin_mma_pmdmxvf16gerx2nn (dmr1024 *, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2NN nothing {dm,pair,dmint,dmr}
+
+ dmr1024 __builtin_mma_pmdmxvf16gerx2nn_internal (dmr1024, v256, vuc, const int<8>, \
+ const int<4>, const int<2>);
+ PMDMXVF16GERX2NN_INTERNAL dmf_pmdmxvf16gerx2nn {dm,pair}
void __builtin_mma_pmdmxvbf16gerx2pn (__dmr1024 *, __vector_pair, vec_t, uint8, uint4, uint2);
void __builtin_mma_pmdmxvbf16gerx2np (__dmr1024 *, __vector_pair, vec_t, uint8, uint4, uint2);
void __builtin_mma_pmdmxvbf16gerx2nn (__dmr1024 *, __vector_pair, vec_t, uint8, uint4, uint2);
+
+void __builtin_mma_dmxvf16gerx2 (__dmr1024 *, __vector_pair, vec_t);
+void __builtin_mma_dmxvf16gerx2pp (__dmr1024 *, __vector_pair, vec_t);
+void __builtin_mma_dmxvf16gerx2pn (__dmr1024 *, __vector_pair, vec_t);
+void __builtin_mma_dmxvf16gerx2np (__dmr1024 *, __vector_pair, vec_t);
+void __builtin_mma_dmxvf16gerx2nn (__dmr1024 *, __vector_pair, vec_t);
+
+void __builtin_mma_pmdmxvf16gerx2 (__dmr1024 *, __vector_pair, vec_t, uint8, uint4, uint2);
+void __builtin_mma_pmdmxvf16gerx2pp (__dmr1024 *, __vector_pair, vec_t, uint8, uint4, uint2);
+void __builtin_mma_pmdmxvf16gerx2pn (__dmr1024 *, __vector_pair, vec_t, uint8, uint4, uint2);
+void __builtin_mma_pmdmxvf16gerx2np (__dmr1024 *, __vector_pair, vec_t, uint8, uint4, uint2);
+void __builtin_mma_pmdmxvf16gerx2nn (__dmr1024 *, __vector_pair, vec_t, uint8, uint4, uint2);
@end smallexample
@node PowerPC Dense Math Facility Built-in Functions
--- /dev/null
+/* { dg-do compile } */
+/* { dg-require-effective-target powerpc_future_compile_ok } */
+/* { dg-options "-mdejagnu-cpu=future -O2" } */
+
+typedef unsigned char vec_t __attribute__((vector_size(16)));
+
+void
+foo (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_dmsetdmrz (&dmr);
+ __builtin_mma_dmxvf16gerx2 (&dmr, vp, vec);
+ *dst = dmr;
+}
+
+void
+bar (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr = dst[0];;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_dmxvf16gerx2 (&dmr, vp, vec);
+ dst[1] = dmr;
+}
+
+/* { dg-final { scan-assembler-times {\mdmxvf16gerx2\M} 2 } } */
+
+void
+foo_1 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_dmsetdmrz (&dmr);
+ __builtin_mma_dmxvf16gerx2nn (&dmr, vp, vec);
+ *dst = dmr;
+}
+
+void
+bar_1 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr = dst[0];;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_dmxvf16gerx2nn (&dmr, vp, vec);
+ dst[1] = dmr;
+}
+
+/* { dg-final { scan-assembler-times {\mdmxvf16gerx2nn\M} 2 } } */
+
+void
+foo_2 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_dmsetdmrz (&dmr);
+ __builtin_mma_dmxvf16gerx2np (&dmr, vp, vec);
+ *dst = dmr;
+}
+
+void
+bar_2 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr = dst[0];;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_dmxvf16gerx2np (&dmr, vp, vec);
+ dst[1] = dmr;
+}
+
+/* { dg-final { scan-assembler-times {\mdmxvf16gerx2np\M} 2 } } */
+
+void
+foo_3 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_dmsetdmrz (&dmr);
+ __builtin_mma_dmxvf16gerx2pn (&dmr, vp, vec);
+ *dst = dmr;
+}
+
+void
+bar_3 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr = dst[0];;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_dmxvf16gerx2pn (&dmr, vp, vec);
+ dst[1] = dmr;
+}
+
+/* { dg-final { scan-assembler-times {\mdmxvf16gerx2pn\M} 2 } } */
+
+void
+foo_4 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_dmsetdmrz (&dmr);
+ __builtin_mma_dmxvf16gerx2pp (&dmr, vp, vec);
+ *dst = dmr;
+}
+
+void
+bar_4 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __dmr1024 dmr = dst[0];;
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_dmxvf16gerx2pp (&dmr, vp, vec);
+ dst[1] = dmr;
+}
+
+/* { dg-final { scan-assembler-times {\mdmxvf16gerx2pp\M} 2 } } */
+
+void
+foo_5 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_pmdmxvf16gerx2 (dst, vp, vec, 255, 15, 2);
+}
+
+/* { dg-final { scan-assembler-times {\mpmdmxvf16gerx2\M} 1 } } */
+
+void
+foo_6 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_pmdmxvf16gerx2nn (dst, vp, vec, 255, 15, 2);
+}
+
+/* { dg-final { scan-assembler-times {\mpmdmxvf16gerx2nn\M} 1 } } */
+
+void
+foo_7 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_pmdmxvf16gerx2np (dst, vp, vec, 255, 15, 2);
+}
+
+/* { dg-final { scan-assembler-times {\mpmdmxvf16gerx2np\M} 1 } } */
+
+void
+foo_8 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_pmdmxvf16gerx2pn (dst, vp, vec, 255, 15, 2);
+}
+
+/* { dg-final { scan-assembler-times {\mpmdmxvf16gerx2pn\M} 1 } } */
+
+void
+foo_9 (__dmr1024 *dst, __vector_pair *vpp, vec_t *src)
+{
+ __vector_pair vp = *vpp;
+ vec_t vec = *src;
+ __builtin_mma_pmdmxvf16gerx2pp (dst, vp, vec, 255, 15, 2);
+}
+
+/* { dg-final { scan-assembler-times {\mpmdmxvf16gerx2pp\M} 1 } } */
projects / thirdparty / gcc.git / commitdiff
