https://gcc.gnu.org/g:b790b7e95d91b5110908c00adcfa0fc392d0f40e
commit b790b7e95d91b5110908c00adcfa0fc392d0f40e Author: Michael Meissner <[email protected]> Date: Thu Jun 25 00:37:44 2026 -0400 Add wD constraint. This is part one of the dense math register patches for the PowerPC. This is the 7th version of the dense math patches. Version 6 of the dense math register patches were posted on April 21st, 2026. * https://gcc.gnu.org/pipermail/gcc-patches/2026-April/713352.html * https://gcc.gnu.org/pipermail/gcc-patches/2026-April/713353.html * https://gcc.gnu.org/pipermail/gcc-patches/2026-April/713354.html * https://gcc.gnu.org/pipermail/gcc-patches/2026-April/713355.html * https://gcc.gnu.org/pipermail/gcc-patches/2026-April/713356.html * https://gcc.gnu.org/pipermail/gcc-patches/2026-April/713357.html This patch needs the -mcpu=future patch posted on April 8th, 2026: * https://gcc.gnu.org/pipermail/gcc-patches/2026-April/712532.html This particular patch did not change from version 6. This patch adds a new constraint ('wD') that matches the accumulator registers used by the MMA instructions. Possible future PowerPC machines are thinking about having a new set of 8 dense math accumulators that will be 1,024 bits in size. The 'wD' constaint was chosen because the VSX constraints start with 'w'. The 'wd' constraint was already used, so I chose 'wD' to be similar. To change code to possibly use dense math registers, the 'd' constraint should be changed to 'wD', and the predicate 'fpr_reg_operand' should be changed to 'accumulator_operand'. On current power10/power11 systems, the accumulators overlap with the 32 traditional FPR registers (i.e. VSX vector registers 0..31). Each accumulator uses 4 adjacent FPR/VSX registers for a 512 bit logical register. Possible future PowerPC machines would have these 8 accumulator registers be separate registers, called dense math registers. It is anticipated that when in dense math register mode, the MMA instructions would use the accumulators instead of the adjacent VSX registers. I.e. in power10/power11 mode, accumulator 1 will overlap with vector registers 4-7, but in dense math register mode, accumulator 1 will be a separate register. Code compiled for power10/power11 systems will continue to work on the potential future machine with dense math register support but the compiler will have fewer vector registers available for allocation because it believe the accumulators are using vector registers. For example, the file mma-double-test.c in the gcc.target/powerpc testsuite directory has 8 more register spills to/from the stack for power10/power11 code then when compiled with dense math register support. I have built bootstrap little endian compilers on power10 systems, and big endian compiler on power9 systems. There were no regression in the tests. Can I add the patches to the GCC trunk after the -mcpu=future patch is applied and GCC 17 has opened up? 2026-06-25 Michael Meissner <[email protected]> gcc/ * config/rs6000/constraints.md (wD): New constraint. * config/rs6000/predicates.md (accumulator_operand): New predicate. * config/rs6000/rs6000.cc (rs6000_debug_reg_global): Print the register class for the 'wD' constraint. (rs6000_init_hard_regno_mode_ok): Set up the 'wD' register constraint class. * config/rs6000/rs6000.h (enum r6000_reg_class_enum): Add element for the 'wD' constraint. * doc/md.texi (PowerPC constraints): Document the 'wD' constraint. Diff: --- gcc/config/rs6000/constraints.md | 3 +++ gcc/config/rs6000/predicates.md | 18 ++++++++++++++++++ gcc/config/rs6000/rs6000.cc | 7 ++++++- gcc/config/rs6000/rs6000.h | 1 + gcc/doc/md.texi | 5 +++++ 5 files changed, 33 insertions(+), 1 deletion(-) diff --git a/gcc/config/rs6000/constraints.md b/gcc/config/rs6000/constraints.md index d0ed47faab84..0d1cde5bd4de 100644 --- a/gcc/config/rs6000/constraints.md +++ b/gcc/config/rs6000/constraints.md @@ -107,6 +107,9 @@ (match_test "TARGET_P8_VECTOR") (match_operand 0 "s5bit_cint_operand"))) +(define_register_constraint "wD" "rs6000_constraints[RS6000_CONSTRAINT_wD]" + "Accumulator register.") + (define_constraint "wE" "@internal Vector constant that can be loaded with the XXSPLTIB instruction." (match_test "xxspltib_constant_nosplit (op, mode)")) diff --git a/gcc/config/rs6000/predicates.md b/gcc/config/rs6000/predicates.md index 4162c22f8f68..7ced624f04e7 100644 --- a/gcc/config/rs6000/predicates.md +++ b/gcc/config/rs6000/predicates.md @@ -186,6 +186,24 @@ return VLOGICAL_REGNO_P (REGNO (op)); }) +;; Return 1 if op is an accumulator. On power10 systems, the accumulators +;; overlap with the FPRs. +(define_predicate "accumulator_operand" + (match_operand 0 "register_operand") +{ + if (SUBREG_P (op)) + op = SUBREG_REG (op); + + if (!REG_P (op)) + return 0; + + if (!HARD_REGISTER_P (op)) + return 1; + + int r = REGNO (op); + return FP_REGNO_P (r) && (r & 3) == 0; +}) + ;; Return 1 if op is the carry register. (define_predicate "ca_operand" (match_operand 0 "register_operand") diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc index d8669d9ffce4..c68b6246242a 100644 --- a/gcc/config/rs6000/rs6000.cc +++ b/gcc/config/rs6000/rs6000.cc @@ -2328,6 +2328,7 @@ rs6000_debug_reg_global (void) "wr reg_class = %s\n" "wx reg_class = %s\n" "wA reg_class = %s\n" + "wD reg_class = %s\n" "\n", reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_d]], reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_v]], @@ -2335,7 +2336,8 @@ rs6000_debug_reg_global (void) reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_we]], reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wr]], reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wx]], - reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wA]]); + reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wA]], + reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wD]]); nl = "\n"; for (m = 0; m < NUM_MACHINE_MODES; ++m) @@ -2992,6 +2994,9 @@ rs6000_init_hard_regno_mode_ok (bool global_init_p) if (TARGET_DIRECT_MOVE_128) rs6000_constraints[RS6000_CONSTRAINT_we] = VSX_REGS; + if (TARGET_MMA) + rs6000_constraints[RS6000_CONSTRAINT_wD] = FLOAT_REGS; + /* Set up the reload helper and direct move functions. */ if (TARGET_VSX || TARGET_ALTIVEC) { diff --git a/gcc/config/rs6000/rs6000.h b/gcc/config/rs6000/rs6000.h index 8235e095bcce..d131144de705 100644 --- a/gcc/config/rs6000/rs6000.h +++ b/gcc/config/rs6000/rs6000.h @@ -1184,6 +1184,7 @@ enum r6000_reg_class_enum { RS6000_CONSTRAINT_wr, /* GPR register if 64-bit */ RS6000_CONSTRAINT_wx, /* FPR register for STFIWX */ RS6000_CONSTRAINT_wA, /* BASE_REGS if 64-bit. */ + RS6000_CONSTRAINT_wD, /* Accumulator regs if MMA/Dense Math. */ RS6000_CONSTRAINT_MAX }; diff --git a/gcc/doc/md.texi b/gcc/doc/md.texi index a98a572cc017..17eb00414506 100644 --- a/gcc/doc/md.texi +++ b/gcc/doc/md.texi @@ -3294,6 +3294,11 @@ Like @code{d}, if @option{-mpowerpc-gfxopt} is used; otherwise, @code{NO_REGS}. @item wA Like @code{b}, if @option{-mpowerpc64} is used; otherwise, @code{NO_REGS}. +@item wD +Accumulator register if @option{-mma} is used; otherwise, +@code{NO_REGS}. For @option{-mcpu=power10} the accumulator registers +overlap with VSX vector registers 0..31. + @item wB Signed 5-bit constant integer that can be loaded into an Altivec register.
