From 0ccb05058312caed86befc75923b6f888ae4e7a6 Mon Sep 17 00:00:00 2001 From: pengxu Date: Tue, 13 May 2025 16:08:33 +0800 Subject: [PATCH 1/2] Loongarch64: fixed cgemm_ncopy_16_lasx --- kernel/loongarch64/cgemm_ncopy_16_lasx.S | 774 +++++++---------------- 1 file changed, 212 insertions(+), 562 deletions(-) diff --git a/kernel/loongarch64/cgemm_ncopy_16_lasx.S b/kernel/loongarch64/cgemm_ncopy_16_lasx.S index 7c2d0ac64..4b9225314 100644 --- a/kernel/loongarch64/cgemm_ncopy_16_lasx.S +++ b/kernel/loongarch64/cgemm_ncopy_16_lasx.S @@ -45,18 +45,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #define S6 $r17 #define S7 $r18 #define S8 $r19 -#define S9 $r20 -#define S10 $r23 -#define S11 $r24 -#define S12 $r25 -#define S13 $r26 -#define S14 $r27 -#define S15 $r28 -#define S16 $r29 -#define TD $r30 -#define TS $r31 +#define S9 $r23 +#define S10 $r24 +#define S11 $r25 +#define S12 $r26 +#define S13 $r27 +#define S14 $r28 +#define S15 $r29 +#define S16 $r30 +#define TD $r20 +#define TS $r11 #define TL $r7 -#define T0 $r6 #define ZERO $r0 #define F0 $f0 @@ -67,6 +66,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #define F5 $f5 #define F6 $f6 #define F7 $f7 +#define F8 $f8 +#define F9 $f9 +#define F10 $f10 +#define F11 $f11 +#define F12 $f12 +#define F13 $f13 +#define F14 $f14 +#define F15 $f15 /* LASX vectors */ #define U0 $xr0 #define U1 $xr1 @@ -103,589 +110,232 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. PROLOGUE - addi.d $sp, $sp, -0x90 - SDARG $r23, $sp, 0x00 - SDARG $r24, $sp, 0x08 - SDARG $r25, $sp, 0x10 - SDARG $r26, $sp, 0x18 - SDARG $r27, $sp, 0x20 - SDARG $r28, $sp, 0x28 - SDARG $r29, $sp, 0x30 - SDARG $r30, $sp, 0x38 - SDARG $r31, $sp, 0x40 - ST $f23, $sp, 0x48 - ST $f24, $sp, 0x50 - ST $f25, $sp, 0x58 - ST $f26, $sp, 0x60 - ST $f27, $sp, 0x68 - ST $f28, $sp, 0x70 - ST $f29, $sp, 0x78 - ST $f30, $sp, 0x80 - ST $f31, $sp, 0x88 - - move TD, DST - move TS, SRC - slli.d TL, LDA, 0x03 - slli.d T0, TL, 0x01 - srai.d J, N, 0x04 + addi.d $sp, $sp, -64 + SDARG $r23, $sp, 0 + SDARG $r24, $sp, 8 + SDARG $r25, $sp, 16 + SDARG $r26, $sp, 24 + SDARG $r27, $sp, 32 + SDARG $r28, $sp, 40 + SDARG $r29, $sp, 48 + SDARG $r30, $sp, 56 + + move TD, DST //boffset + move TS, SRC //aoffset + slli.d TL, LDA, 0x03 //lda + srai.d J, N, 0x04 //j beq J, ZERO, .L_N8 -.L_J1: /* J-- */ +.L_J1: /* if(j>0) j--*/ move S1, TS add.d S2, TS, TL - srai.d I, M, 0x03 + move I, M add.d S3, S2, TL - addi.d J, J, -1 add.d S4, S3, TL - add.d S5, S3, T0 - add.d S6, S4, T0 - add.d S7, S5, T0 - add.d S8, S6, T0 - add.d S9, S7, T0 - add.d S10, S8, T0 - add.d S11, S9, T0 - add.d S12, S10, T0 - add.d S13, S11, T0 - add.d S14, S12, T0 - add.d S15, S13, T0 - add.d S16, S14, T0 - add.d TS, S15, T0 - beq I, ZERO, .L_I7 - -.L_I1: /* I-- */ - xvld U0, S1, 0x00 - xvld U1, S2, 0x00 - xvld U2, S3, 0x00 - xvld U3, S4, 0x00 - xvld U4, S5, 0x00 - xvld U5, S6, 0x00 - xvld U6, S7, 0x00 - xvld U7, S8, 0x00 - xvld U8, S9, 0x00 - xvld U9, S10, 0x00 - xvld U10, S11, 0x00 - xvld U11, S12, 0x00 - xvld U12, S13, 0x00 - xvld U13, S14, 0x00 - xvld U14, S15, 0x00 - xvld U15, S16, 0x00 - - xvpackev.d D0, U1, U0 - xvpackod.d D1, U1, U0 - xvpackev.d D2, U3, U2 - xvpackod.d D3, U3, U2 - xvpackev.d D4, U5, U4 - xvpackod.d D5, U5, U4 - xvpackev.d D6, U7, U6 - xvpackod.d D7, U7, U6 - - xvpackev.d D8, U9, U8 - xvpackod.d D9, U9, U8 - xvpackev.d D10, U11, U10 - xvpackod.d D11, U11, U10 - xvpackev.d D12, U13, U12 - xvpackod.d D13, U13, U12 - xvpackev.d D14, U15, U14 - xvpackod.d D15, U15, U14 - - xvand.v U0, D0, D0 - xvpermi.q D0, D2, 0x02 // 0 - xvand.v U4, D4, D4 - xvpermi.q D4, D6, 0x02 // 1 - xvand.v U1, D1, D1 - xvpermi.q D1, D3, 0x02 // 4 - xvand.v U5, D5, D5 - xvpermi.q D5, D7, 0x02 // 5 - xvpermi.q D2, U0, 0x31 // 8 - xvpermi.q D6, U4, 0x31 // 9 - xvpermi.q D3, U1, 0x31 // 12 - xvpermi.q D7, U5, 0x31 // 13 - - xvand.v U8, D8, D8 - xvpermi.q D8, D10, 0x02 // 2 - xvand.v U12, D12, D12 - xvpermi.q D12, D14, 0x02 // 3 - xvand.v U9, D9, D9 - xvpermi.q D9, D11, 0x02 // 6 - xvand.v U13, D13, D13 - xvpermi.q D13, D15, 0x02 // 7 - xvpermi.q D10, U8, 0x31 // 10 - xvpermi.q D14, U12, 0x31 // 11 - xvpermi.q D11, U9, 0x31 // 14 - xvpermi.q D15, U13, 0x31 // 15 - - xvst D0, TD, 0x00 // 0 - xvst D4, TD, 0x20 // 1 - xvst D8, TD, 0x40 // 2 - xvst D12, TD, 0x60 // 3 - xvst D1, TD, 0x80 // 4 - xvst D5, TD, 0xA0 // 5 - xvst D9, TD, 0xC0 // 6 - xvst D13, TD, 0xE0 // 7 - addi.d TD, TD, 0x100 - xvst D2, TD, 0x00 // 8 - xvst D6, TD, 0x20 // 9 - xvst D10, TD, 0x40 // 10 - xvst D14, TD, 0x60 // 11 - xvst D3, TD, 0x80 // 12 - xvst D7, TD, 0xA0 // 13 - xvst D11, TD, 0xC0 // 14 - xvst D15, TD, 0xE0 // 15 - addi.d TD, TD, 0x100 - - xvld U0, S1, 0x20 - xvld U1, S2, 0x20 - xvld U2, S3, 0x20 - xvld U3, S4, 0x20 - xvld U4, S5, 0x20 - xvld U5, S6, 0x20 - xvld U6, S7, 0x20 - xvld U7, S8, 0x20 - xvld U8, S9, 0x20 - xvld U9, S10, 0x20 - xvld U10, S11, 0x20 - xvld U11, S12, 0x20 - xvld U12, S13, 0x20 - xvld U13, S14, 0x20 - xvld U14, S15, 0x20 - xvld U15, S16, 0x20 - - xvpackev.d D0, U1, U0 - xvpackod.d D1, U1, U0 - xvpackev.d D2, U3, U2 - xvpackod.d D3, U3, U2 - xvpackev.d D4, U5, U4 - xvpackod.d D5, U5, U4 - xvpackev.d D6, U7, U6 - xvpackod.d D7, U7, U6 - - xvpackev.d D8, U9, U8 - xvpackod.d D9, U9, U8 - xvpackev.d D10, U11, U10 - xvpackod.d D11, U11, U10 - xvpackev.d D12, U13, U12 - xvpackod.d D13, U13, U12 - xvpackev.d D14, U15, U14 - xvpackod.d D15, U15, U14 - - xvand.v U0, D0, D0 - xvpermi.q D0, D2, 0x02 // 0 - xvand.v U4, D4, D4 - xvpermi.q D4, D6, 0x02 // 1 - xvand.v U1, D1, D1 - xvpermi.q D1, D3, 0x02 // 4 - xvand.v U5, D5, D5 - xvpermi.q D5, D7, 0x02 // 5 - xvpermi.q D2, U0, 0x31 // 8 - xvpermi.q D6, U4, 0x31 // 9 - xvpermi.q D3, U1, 0x31 // 12 - xvpermi.q D7, U5, 0x31 // 13 - - xvand.v U8, D8, D8 - xvpermi.q D8, D10, 0x02 // 2 - xvand.v U12, D12, D12 - xvpermi.q D12, D14, 0x02 // 3 - xvand.v U9, D9, D9 - xvpermi.q D9, D11, 0x02 // 6 - xvand.v U13, D13, D13 - xvpermi.q D13, D15, 0x02 // 7 - xvpermi.q D10, U8, 0x31 // 10 - xvpermi.q D14, U12, 0x31 // 11 - xvpermi.q D11, U9, 0x31 // 14 - xvpermi.q D15, U13, 0x31 // 15 - - xvst D0, TD, 0x00 // 0 - xvst D4, TD, 0x20 // 1 - xvst D8, TD, 0x40 // 2 - xvst D12, TD, 0x60 // 3 - xvst D1, TD, 0x80 // 4 - xvst D5, TD, 0xA0 // 5 - xvst D9, TD, 0xC0 // 6 - xvst D13, TD, 0xE0 // 7 - addi.d TD, TD, 0x100 - xvst D2, TD, 0x00 // 8 - xvst D6, TD, 0x20 // 9 - xvst D10, TD, 0x40 // 10 - xvst D14, TD, 0x60 // 11 - xvst D3, TD, 0x80 // 12 - xvst D7, TD, 0xA0 // 13 - xvst D11, TD, 0xC0 // 14 - xvst D15, TD, 0xE0 // 15 - addi.d TD, TD, 0x100 - - - addi.d S1, S1, 0x40 - addi.d S2, S2, 0x40 - addi.d S3, S3, 0x40 - addi.d S4, S4, 0x40 - addi.d S5, S5, 0x40 - addi.d S6, S6, 0x40 - addi.d S7, S7, 0x40 - addi.d S8, S8, 0x40 - addi.d S9, S9, 0x40 - addi.d S10, S10, 0x40 - addi.d S11, S11, 0x40 - addi.d S12, S12, 0x40 - addi.d S13, S13, 0x40 - addi.d S14, S14, 0x40 - addi.d S15, S15, 0x40 - addi.d S16, S16, 0x40 - + add.d S5, S4, TL + add.d S6, S5, TL + add.d S7, S6, TL + add.d S8, S7, TL + add.d S9, S8, TL + add.d S10, S9, TL + add.d S11, S10, TL + add.d S12, S11, TL + add.d S13, S12, TL + add.d S14, S13, TL + add.d S15, S14, TL + add.d S16, S15, TL + add.d TS, S16, TL + beq I, ZERO, .L_J11 + +.L_I1: /* if(i>0) i--*/ + fld.d F0, S1, 0x00 + fld.d F1, S2, 0x00 + fld.d F2, S3, 0x00 + fld.d F3, S4, 0x00 + fld.d F4, S5, 0x00 + fld.d F5, S6, 0x00 + fld.d F6, S7, 0x00 + fld.d F7, S8, 0x00 + + fst.d F0, TD, 0x00 + fst.d F1, TD, 0x08 + fst.d F2, TD, 0x10 + fst.d F3, TD, 0x18 + fst.d F4, TD, 0x20 + fst.d F5, TD, 0x28 + fst.d F6, TD, 0x30 + fst.d F7, TD, 0x38 + + fld.d F0, S9, 0x00 + fld.d F1, S10, 0x00 + fld.d F2, S11, 0x00 + fld.d F3, S12, 0x00 + fld.d F4, S13, 0x00 + fld.d F5, S14, 0x00 + fld.d F6, S15, 0x00 + fld.d F7, S16, 0x00 + + fst.d F0, TD, 0x40 + fst.d F1, TD, 0x48 + fst.d F2, TD, 0x50 + fst.d F3, TD, 0x58 + fst.d F4, TD, 0x60 + fst.d F5, TD, 0x68 + fst.d F6, TD, 0x70 + fst.d F7, TD, 0x78 + + addi.d S1, S1, 0x08 + addi.d S2, S2, 0x08 + addi.d S3, S3, 0x08 + addi.d S4, S4, 0x08 + addi.d S5, S5, 0x08 + addi.d S6, S6, 0x08 + addi.d S7, S7, 0x08 + addi.d S8, S8, 0x08 + addi.d S9, S9, 0x08 + addi.d S10, S10, 0x08 + addi.d S11, S11, 0x08 + addi.d S12, S12, 0x08 + addi.d S13, S13, 0x08 + addi.d S14, S14, 0x08 + addi.d S15, S15, 0x08 + addi.d S16, S16, 0x08 + addi.d TD, TD, 0x80 addi.d I, I, -1 blt ZERO, I, .L_I1 -.L_I7: - andi I, M, 0x07 - beq I, ZERO, .L_I0 - -.L_II1: /* I-- */ - fld.d F0, S1, 0x00 - fld.d F1, S2, 0x00 - fld.d F2, S3, 0x00 - fld.d F3, S4, 0x00 - fld.d F4, S5, 0x00 - fld.d F5, S6, 0x00 - fld.d F6, S7, 0x00 - fld.d F7, S8, 0x00 - - fst.d F0, TD, 0x00 - addi.d S1, S1, 0x08 - fst.d F1, TD, 0x08 - addi.d S2, S2, 0x08 - fst.d F2, TD, 0x10 - addi.d S3, S3, 0x08 - fst.d F3, TD, 0x18 - addi.d S4, S4, 0x08 - fst.d F4, TD, 0x20 - addi.d S5, S5, 0x08 - fst.d F5, TD, 0x28 - addi.d S6, S6, 0x08 - fst.d F6, TD, 0x30 - addi.d S7, S7, 0x08 - fst.d F7, TD, 0x38 - addi.d S8, S8, 0x08 - addi.d TD, TD, 0x40 - - fld.d F0, S9, 0x00 - fld.d F1, S10, 0x00 - fld.d F2, S11, 0x00 - fld.d F3, S12, 0x00 - fld.d F4, S13, 0x00 - fld.d F5, S14, 0x00 - fld.d F6, S15, 0x00 - fld.d F7, S16, 0x00 - - fst.d F0, TD, 0x00 - addi.d S9, S9, 0x08 - fst.d F1, TD, 0x08 - addi.d S10, S10, 0x08 - fst.d F2, TD, 0x10 - addi.d S11, S11, 0x08 - fst.d F3, TD, 0x18 - addi.d S12, S12, 0x08 - fst.d F4, TD, 0x20 - addi.d S13, S13, 0x08 - fst.d F5, TD, 0x28 - addi.d S14, S14, 0x08 - fst.d F6, TD, 0x30 - addi.d S15, S15, 0x08 - fst.d F7, TD, 0x38 - addi.d S16, S16, 0x08 - addi.d TD, TD, 0x40 - - addi.d I, I, -1 - blt ZERO, I, .L_II1 - -.L_I0: - blt ZERO, J, .L_J1 - -.L_N8: - andi J, N, 0x08 - beq ZERO, J, .L_N4 +.L_J11: /* j--*/ + addi.d J, J, -1 + blt ZERO, J, .L_J1 + +.L_N8: /* if(n&8)*/ + andi I, N, 0x08 + beq I, ZERO, .L_N4 move S1, TS add.d S2, TS, TL - srai.d I, M, 0x03 + move I, M add.d S3, S2, TL - add.d S4, S2, T0 - add.d S5, S3, T0 - add.d S6, S4, T0 - add.d S7, S5, T0 - add.d S8, S6, T0 - add.d TS, S7, T0 - beq I, ZERO, .L_8I3 - -.L_8I1: /* I-- */ - xvld U0, S1, 0x00 - xvld U1, S2, 0x00 - xvld U2, S3, 0x00 - xvld U3, S4, 0x00 - xvld U4, S5, 0x00 - xvld U5, S6, 0x00 - xvld U6, S7, 0x00 - xvld U7, S8, 0x00 - - xvpackev.d D0, U1, U0 - xvpackod.d D1, U1, U0 - xvpackev.d D2, U3, U2 - xvpackod.d D3, U3, U2 - xvpackev.d D4, U5, U4 - xvpackod.d D5, U5, U4 - xvpackev.d D6, U7, U6 - xvpackod.d D7, U7, U6 - - xvand.v U0, D0, D0 - xvpermi.q D0, D2, 0x02 // 0 - xvand.v U4, D4, D4 - xvpermi.q D4, D6, 0x02 // 1 - xvand.v U1, D1, D1 - xvpermi.q D1, D3, 0x02 // 2 - xvand.v U5, D5, D5 - xvpermi.q D5, D7, 0x02 // 3 - xvpermi.q D2, U0, 0x31 // 4 - xvpermi.q D6, U4, 0x31 // 5 - xvpermi.q D3, U1, 0x31 // 6 - xvpermi.q D7, U5, 0x31 // 7 - - xvst D0, TD, 0x00 - xvst D4, TD, 0x20 - xvst D1, TD, 0x40 - xvst D5, TD, 0x60 - xvst D2, TD, 0x80 - xvst D6, TD, 0xA0 - xvst D3, TD, 0xC0 - xvst D7, TD, 0xE0 - addi.d TD, TD, 0x100 - - xvld U0, S1, 0x20 - xvld U1, S2, 0x20 - xvld U2, S3, 0x20 - xvld U3, S4, 0x20 - xvld U4, S5, 0x20 - xvld U5, S6, 0x20 - xvld U6, S7, 0x20 - xvld U7, S8, 0x20 - - xvpackev.d D0, U1, U0 - xvpackod.d D1, U1, U0 - xvpackev.d D2, U3, U2 - xvpackod.d D3, U3, U2 - xvpackev.d D4, U5, U4 - xvpackod.d D5, U5, U4 - xvpackev.d D6, U7, U6 - xvpackod.d D7, U7, U6 - - xvand.v U0, D0, D0 - xvpermi.q D0, D2, 0x02 // 0 - xvand.v U4, D4, D4 - xvpermi.q D4, D6, 0x02 // 1 - xvand.v U1, D1, D1 - xvpermi.q D1, D3, 0x02 // 2 - xvand.v U5, D5, D5 - xvpermi.q D5, D7, 0x02 // 3 - xvpermi.q D2, U0, 0x31 // 4 - xvpermi.q D6, U4, 0x31 // 5 - xvpermi.q D3, U1, 0x31 // 6 - xvpermi.q D7, U5, 0x31 // 7 - - xvst D0, TD, 0x00 - xvst D4, TD, 0x20 - xvst D1, TD, 0x40 - xvst D5, TD, 0x60 - xvst D2, TD, 0x80 - xvst D6, TD, 0xA0 - xvst D3, TD, 0xC0 - xvst D7, TD, 0xE0 - addi.d TD, TD, 0x100 - - addi.d S1, S1, 0x40 - addi.d S2, S2, 0x40 - addi.d S3, S3, 0x40 - addi.d S4, S4, 0x40 - addi.d S5, S5, 0x40 - addi.d S6, S6, 0x40 - addi.d S7, S7, 0x40 - addi.d S8, S8, 0x40 - + add.d S4, S3, TL + add.d S5, S4, TL + add.d S6, S5, TL + add.d S7, S6, TL + add.d S8, S7, TL + add.d TS, S8, TL + beq I, ZERO, .L_N4 + +.L_N81: /* if(i>0) i--*/ + fld.d F0, S1, 0x00 + fld.d F1, S2, 0x00 + fld.d F2, S3, 0x00 + fld.d F3, S4, 0x00 + fld.d F4, S5, 0x00 + fld.d F5, S6, 0x00 + fld.d F6, S7, 0x00 + fld.d F7, S8, 0x00 + + fst.d F0, TD, 0x00 + fst.d F1, TD, 0x08 + fst.d F2, TD, 0x10 + fst.d F3, TD, 0x18 + fst.d F4, TD, 0x20 + fst.d F5, TD, 0x28 + fst.d F6, TD, 0x30 + fst.d F7, TD, 0x38 + + addi.d S1, S1, 0x08 + addi.d S2, S2, 0x08 + addi.d S3, S3, 0x08 + addi.d S4, S4, 0x08 + addi.d S5, S5, 0x08 + addi.d S6, S6, 0x08 + addi.d S7, S7, 0x08 + addi.d S8, S8, 0x08 + addi.d TD, TD, 0x40 addi.d I, I, -1 - blt ZERO, I, .L_8I1 - -.L_8I3: - andi I, M, 0x07 - beq I, ZERO, .L_N4 - -.L_8I11: - fld.d F0, S1, 0x00 - fld.d F1, S2, 0x00 - fld.d F2, S3, 0x00 - fld.d F3, S4, 0x00 - fld.d F4, S5, 0x00 - fld.d F5, S6, 0x00 - fld.d F6, S7, 0x00 - fld.d F7, S8, 0x00 - - fst.d F0, TD, 0x00 - addi.d S1, S1, 0x08 - fst.d F1, TD, 0x08 - addi.d S2, S2, 0x08 - fst.d F2, TD, 0x10 - addi.d S3, S3, 0x08 - fst.d F3, TD, 0x18 - addi.d S4, S4, 0x08 - fst.d F4, TD, 0x20 - addi.d S5, S5, 0x08 - fst.d F5, TD, 0x28 - addi.d S6, S6, 0x08 - fst.d F6, TD, 0x30 - addi.d S7, S7, 0x08 - fst.d F7, TD, 0x38 - addi.d S8, S8, 0x08 - - addi.d TD, TD, 0x40 - addi.d I, I, -1 - blt ZERO, I, .L_8I11 - -.L_N4: - andi J, N, 0x04 - beq ZERO, J, .L_N2 + blt ZERO, I, .L_N81 + +.L_N4: /* if(n&4)*/ + andi I, N, 0x04 + beq I, ZERO, .L_N2 move S1, TS add.d S2, TS, TL - srai.d I, M, 0x02 + move I, M add.d S3, S2, TL - add.d S4, S2, T0 - add.d TS, S3, T0 - beq I, ZERO, .L_I3 - -.L_4I1: /* I-- */ - xvld U0, S1, 0x00 - xvld U1, S2, 0x00 - xvld U2, S3, 0x00 - xvld U3, S4, 0x00 - - xvpackev.d D0, U1, U0 - xvpackod.d D1, U1, U0 - xvpackev.d D2, U3, U2 - xvpackod.d D3, U3, U2 - - xvand.v U0, D0, D0 - xvpermi.q D0, D2, 0x02 // 0 - xvand.v U1, D1, D1 - xvpermi.q D1, D3, 0x02 // 1 - xvpermi.q D2, U0, 0x31 // 2 - xvpermi.q D3, U1, 0x31 // 3 - - xvst D0, TD, 0x00 - xvst D1, TD, 0x20 - xvst D2, TD, 0x40 - xvst D3, TD, 0x60 - - addi.d S1, S1, 0x20 - addi.d S2, S2, 0x20 - addi.d S3, S3, 0x20 - addi.d S4, S4, 0x20 - addi.d TD, TD, 0x80 - + add.d S4, S3, TL + add.d TS, S4, TL + beq I, ZERO, .L_N2 + +.L_N41: /* if(i>0)*/ + fld.d F0, S1, 0x00 + fld.d F1, S2, 0x00 + fld.d F2, S3, 0x00 + fld.d F3, S4, 0x00 + + fst.d F0, TD, 0x00 + fst.d F1, TD, 0x08 + fst.d F2, TD, 0x10 + fst.d F3, TD, 0x18 + + addi.d S1, S1, 0x08 + addi.d S2, S2, 0x08 + addi.d S3, S3, 0x08 + addi.d S4, S4, 0x08 + addi.d TD, TD, 0x20 addi.d I, I, -1 - blt ZERO, I, .L_4I1 - -.L_I3: - andi I, M, 0x03 - beq I, ZERO, .L_N2 - -.L_4II1: - fld.d F0, S1, 0x00 - fld.d F1, S2, 0x00 - fld.d F2, S3, 0x00 - fld.d F3, S4, 0x00 - - fst.d F0, TD, 0x00 - addi.d S1, S1, 0x08 - fst.d F1, TD, 0x08 - addi.d S2, S2, 0x08 - fst.d F2, TD, 0x10 - addi.d S3, S3, 0x08 - fst.d F3, TD, 0x18 - addi.d S4, S4, 0x08 - - addi.d TD, TD, 0x20 - addi.d I, I, -1 - blt ZERO, I, .L_4II1 - -.L_N2: - andi J, N, 0x02 - beq ZERO, J, .L_N1 + blt ZERO, I, .L_N41 + +.L_N2: /* if(n&2)*/ + andi I, N, 0x02 + beq I, ZERO, .L_N1 move S1, TS add.d S2, TS, TL - srai.d I, M, 0x01 + move I, M add.d TS, S2, TL - beq I, ZERO, .L_NI1 - -.L_2I1: /* I-- */ - xvld U0, S1, 0x00 - xvld U1, S2, 0x00 - - xvpackev.d D0, U1, U0 - xvpackod.d D1, U1, U0 - - xvpermi.q D0, D1, 0x02 // 0 + beq I, ZERO, .L_N1 - xvst D0, TD, 0x00 +.L_N21: /* if(i>0)*/ + fld.d F0, S1, 0x00 + fld.d F1, S2, 0x00 - addi.d S1, S1, 0x10 - addi.d S2, S2, 0x10 - addi.d TD, TD, 0x20 + fst.d F0, TD, 0x00 + fst.d F1, TD, 0x08 + addi.d S1, S1, 0x08 + addi.d S2, S2, 0x08 + addi.d TD, TD, 0x10 addi.d I, I, -1 - blt ZERO, I, .L_2I1 - -.L_NI1: - andi I, M, 0x01 - beq I, ZERO, .L_N1 - + blt ZERO, I, .L_N21 - fld.d F0, S1, 0x00 - fld.d F1, S2, 0x00 +.L_N1: /* if(n&2)*/ + andi I, N, 0x01 + beq I, ZERO, .L_N0 - fst.d F0, TD, 0x00 - addi.d S1, S1, 0x08 - fst.d F1, TD, 0x08 - addi.d S2, S2, 0x08 - addi.d TD, TD, 0x10 + move S1, TS + move I, M + beq I, ZERO, .L_N0 -.L_N1: - move S1, TS - beq ZERO, M, .L_N0 +.L_N11: /* if(i>0)*/ + fld.d F0, S1, 0x00 + fst.d F0, TD, 0x00 -.L_M1: - fld.d F0, S1, 0x00 - addi.d S1, S1, 0x08 - fst.d F0, TD, 0x00 - addi.d TD, TD, 0x08 - addi.d M, M, -1 - blt ZERO, M, .L_M1 + addi.d S1, S1, 0x08 + addi.d TD, TD, 0x08 + addi.d I, I, -1 + blt ZERO, I, .L_N11 .L_N0: - LDARG $r23, $sp, 0x00 - LDARG $r24, $sp, 0x08 - LDARG $r25, $sp, 0x10 - LDARG $r26, $sp, 0x18 - LDARG $r27, $sp, 0x20 - LDARG $r28, $sp, 0x28 - LDARG $r29, $sp, 0x30 - LDARG $r30, $sp, 0x38 - LDARG $r31, $sp, 0x40 - LD $f23, $sp, 0x48 - LD $f24, $sp, 0x50 - LD $f25, $sp, 0x58 - LD $f26, $sp, 0x60 - LD $f27, $sp, 0x68 - LD $f28, $sp, 0x70 - LD $f29, $sp, 0x78 - LD $f30, $sp, 0x80 - LD $f31, $sp, 0x88 - addi.d $sp, $sp, 0x90 - jirl $r0, $r1, 0x00 + LDARG $r23, $sp, 0 + LDARG $r24, $sp, 8 + LDARG $r25, $sp, 16 + LDARG $r26, $sp, 24 + LDARG $r27, $sp, 32 + LDARG $r28, $sp, 40 + LDARG $r29, $sp, 48 + LDARG $r30, $sp, 56 + addi.d $sp, $sp, 64 + jirl $r0, $r1, 0x00 EPILOGUE \ No newline at end of file From a978ad318070e24f5ca0cf4221b55abb5869287e Mon Sep 17 00:00:00 2001 From: pengxu Date: Tue, 13 May 2025 16:09:12 +0800 Subject: [PATCH 2/2] Loongarch64: add C functions of zgemm_ncopy_16 --- kernel/generic/zgemm_ncopy_16.c | 332 ++++++++++++++++++++++++++++++++ 1 file changed, 332 insertions(+) create mode 100644 kernel/generic/zgemm_ncopy_16.c diff --git a/kernel/generic/zgemm_ncopy_16.c b/kernel/generic/zgemm_ncopy_16.c new file mode 100644 index 000000000..088103525 --- /dev/null +++ b/kernel/generic/zgemm_ncopy_16.c @@ -0,0 +1,332 @@ +/*********************************************************************/ +/* Copyright 2009, 2010 The University of Texas at Austin. */ +/* All rights reserved. */ +/* */ +/* Redistribution and use in source and binary forms, with or */ +/* without modification, are permitted provided that the following */ +/* conditions are met: */ +/* */ +/* 1. Redistributions of source code must retain the above */ +/* copyright notice, this list of conditions and the following */ +/* disclaimer. */ +/* */ +/* 2. Redistributions in binary form must reproduce the above */ +/* copyright notice, this list of conditions and the following */ +/* disclaimer in the documentation and/or other materials */ +/* provided with the distribution. */ +/* */ +/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */ +/* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */ +/* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */ +/* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */ +/* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */ +/* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */ +/* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */ +/* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */ +/* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */ +/* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */ +/* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */ +/* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */ +/* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */ +/* POSSIBILITY OF SUCH DAMAGE. */ +/* */ +/* The views and conclusions contained in the software and */ +/* documentation are those of the authors and should not be */ +/* interpreted as representing official policies, either expressed */ +/* or implied, of The University of Texas at Austin. */ +/*********************************************************************/ + +#include +#include "common.h" + +int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){ + BLASLONG i, j; + + IFLOAT *aoffset; + IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4; + IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8; + IFLOAT *aoffset9, *aoffset10, *aoffset11, *aoffset12; + IFLOAT *aoffset13, *aoffset14, *aoffset15, *aoffset16; + + IFLOAT *boffset; + IFLOAT ctemp01, ctemp02, ctemp03, ctemp04; + IFLOAT ctemp05, ctemp06, ctemp07, ctemp08; + IFLOAT ctemp09, ctemp10, ctemp11, ctemp12; + IFLOAT ctemp13, ctemp14, ctemp15, ctemp16; + IFLOAT ctemp17, ctemp18, ctemp19, ctemp20; + IFLOAT ctemp21, ctemp22, ctemp23, ctemp24; + IFLOAT ctemp25, ctemp26, ctemp27, ctemp28; + IFLOAT ctemp29, ctemp30, ctemp31, ctemp32; + + aoffset = a; + boffset = b; + lda *= 2; + + j = (n >> 4); + if (j > 0){ + do{ + aoffset1 = aoffset; + aoffset2 = aoffset1 + lda; + aoffset3 = aoffset2 + lda; + aoffset4 = aoffset3 + lda; + aoffset5 = aoffset4 + lda; + aoffset6 = aoffset5 + lda; + aoffset7 = aoffset6 + lda; + aoffset8 = aoffset7 + lda; + aoffset9 = aoffset8 + lda; + aoffset10 = aoffset9 + lda; + aoffset11 = aoffset10 + lda; + aoffset12 = aoffset11 + lda; + aoffset13 = aoffset12 + lda; + aoffset14 = aoffset13 + lda; + aoffset15 = aoffset14 + lda; + aoffset16 = aoffset15 + lda; + aoffset += 16 * lda; + + i = m; + if (i > 0){ + do{ + ctemp01 = *(aoffset1 + 0); + ctemp02 = *(aoffset1 + 1); + ctemp03 = *(aoffset2 + 0); + ctemp04 = *(aoffset2 + 1); + ctemp05 = *(aoffset3 + 0); + ctemp06 = *(aoffset3 + 1); + ctemp07 = *(aoffset4 + 0); + ctemp08 = *(aoffset4 + 1); + ctemp09 = *(aoffset5 + 0); + ctemp10 = *(aoffset5 + 1); + ctemp11 = *(aoffset6 + 0); + ctemp12 = *(aoffset6 + 1); + ctemp13 = *(aoffset7 + 0); + ctemp14 = *(aoffset7 + 1); + ctemp15 = *(aoffset8 + 0); + ctemp16 = *(aoffset8 + 1); + + ctemp17 = *(aoffset9 + 0); + ctemp18 = *(aoffset9 + 1); + ctemp19 = *(aoffset10 + 0); + ctemp20 = *(aoffset10 + 1); + ctemp21 = *(aoffset11 + 0); + ctemp22 = *(aoffset11 + 1); + ctemp23 = *(aoffset12 + 0); + ctemp24 = *(aoffset12 + 1); + ctemp25 = *(aoffset13 + 0); + ctemp26 = *(aoffset13 + 1); + ctemp27 = *(aoffset14 + 0); + ctemp28 = *(aoffset14 + 1); + ctemp29 = *(aoffset15 + 0); + ctemp30 = *(aoffset15 + 1); + ctemp31 = *(aoffset16 + 0); + ctemp32 = *(aoffset16 + 1); + + *(boffset + 0) = ctemp01; + *(boffset + 1) = ctemp02; + *(boffset + 2) = ctemp03; + *(boffset + 3) = ctemp04; + *(boffset + 4) = ctemp05; + *(boffset + 5) = ctemp06; + *(boffset + 6) = ctemp07; + *(boffset + 7) = ctemp08; + *(boffset + 8) = ctemp09; + *(boffset + 9) = ctemp10; + *(boffset + 10) = ctemp11; + *(boffset + 11) = ctemp12; + *(boffset + 12) = ctemp13; + *(boffset + 13) = ctemp14; + *(boffset + 14) = ctemp15; + *(boffset + 15) = ctemp16; + + *(boffset + 16) = ctemp17; + *(boffset + 17) = ctemp18; + *(boffset + 18) = ctemp19; + *(boffset + 19) = ctemp20; + *(boffset + 20) = ctemp21; + *(boffset + 21) = ctemp22; + *(boffset + 22) = ctemp23; + *(boffset + 23) = ctemp24; + *(boffset + 24) = ctemp25; + *(boffset + 25) = ctemp26; + *(boffset + 26) = ctemp27; + *(boffset + 27) = ctemp28; + *(boffset + 28) = ctemp29; + *(boffset + 29) = ctemp30; + *(boffset + 30) = ctemp31; + *(boffset + 31) = ctemp32; + + aoffset1 += 2; + aoffset2 += 2; + aoffset3 += 2; + aoffset4 += 2; + aoffset5 += 2; + aoffset6 += 2; + aoffset7 += 2; + aoffset8 += 2; + aoffset9 += 2; + aoffset10 += 2; + aoffset11 += 2; + aoffset12 += 2; + aoffset13 += 2; + aoffset14 += 2; + aoffset15 += 2; + aoffset16 += 2; + + boffset += 32; + i --; + }while(i > 0); + } + j--; + }while(j > 0); + } /* end of if(j > 0) */ + + if (n & 8){ + aoffset1 = aoffset; + aoffset2 = aoffset1 + lda; + aoffset3 = aoffset2 + lda; + aoffset4 = aoffset3 + lda; + aoffset5 = aoffset4 + lda; + aoffset6 = aoffset5 + lda; + aoffset7 = aoffset6 + lda; + aoffset8 = aoffset7 + lda; + aoffset += 8 * lda; + + i = m; + if (i > 0){ + do{ + ctemp01 = *(aoffset1 + 0); + ctemp02 = *(aoffset1 + 1); + ctemp03 = *(aoffset2 + 0); + ctemp04 = *(aoffset2 + 1); + ctemp05 = *(aoffset3 + 0); + ctemp06 = *(aoffset3 + 1); + ctemp07 = *(aoffset4 + 0); + ctemp08 = *(aoffset4 + 1); + ctemp09 = *(aoffset5 + 0); + ctemp10 = *(aoffset5 + 1); + ctemp11 = *(aoffset6 + 0); + ctemp12 = *(aoffset6 + 1); + ctemp13 = *(aoffset7 + 0); + ctemp14 = *(aoffset7 + 1); + ctemp15 = *(aoffset8 + 0); + ctemp16 = *(aoffset8 + 1); + + *(boffset + 0) = ctemp01; + *(boffset + 1) = ctemp02; + *(boffset + 2) = ctemp03; + *(boffset + 3) = ctemp04; + *(boffset + 4) = ctemp05; + *(boffset + 5) = ctemp06; + *(boffset + 6) = ctemp07; + *(boffset + 7) = ctemp08; + *(boffset + 8) = ctemp09; + *(boffset + 9) = ctemp10; + *(boffset + 10) = ctemp11; + *(boffset + 11) = ctemp12; + *(boffset + 12) = ctemp13; + *(boffset + 13) = ctemp14; + *(boffset + 14) = ctemp15; + *(boffset + 15) = ctemp16; + + aoffset1 += 2; + aoffset2 += 2; + aoffset3 += 2; + aoffset4 += 2; + aoffset5 += 2; + aoffset6 += 2; + aoffset7 += 2; + aoffset8 += 2; + + boffset += 16; + i --; + }while(i > 0); + } + } + + if (n & 4){ + aoffset1 = aoffset; + aoffset2 = aoffset1 + lda; + aoffset3 = aoffset2 + lda; + aoffset4 = aoffset3 + lda; + aoffset += 4 * lda; + + i = m; + if (i > 0){ + do{ + ctemp01 = *(aoffset1 + 0); + ctemp02 = *(aoffset1 + 1); + ctemp03 = *(aoffset2 + 0); + ctemp04 = *(aoffset2 + 1); + ctemp05 = *(aoffset3 + 0); + ctemp06 = *(aoffset3 + 1); + ctemp07 = *(aoffset4 + 0); + ctemp08 = *(aoffset4 + 1); + + *(boffset + 0) = ctemp01; + *(boffset + 1) = ctemp02; + *(boffset + 2) = ctemp03; + *(boffset + 3) = ctemp04; + *(boffset + 4) = ctemp05; + *(boffset + 5) = ctemp06; + *(boffset + 6) = ctemp07; + *(boffset + 7) = ctemp08; + + aoffset1 += 2; + aoffset2 += 2; + aoffset3 += 2; + aoffset4 += 2; + + boffset += 8; + i --; + }while(i > 0); + } + } /* end of if(j > 0) */ + + if (n & 2){ + aoffset1 = aoffset; + aoffset2 = aoffset1 + lda; + aoffset += 2 * lda; + + i = m; + if (i > 0){ + do{ + ctemp01 = *(aoffset1 + 0); + ctemp02 = *(aoffset1 + 1); + ctemp03 = *(aoffset2 + 0); + ctemp04 = *(aoffset2 + 1); + + *(boffset + 0) = ctemp01; + *(boffset + 1) = ctemp02; + *(boffset + 2) = ctemp03; + *(boffset + 3) = ctemp04; + + aoffset1 += 2; + aoffset2 += 2; + boffset += 4; + i --; + }while(i > 0); + } + + } /* end of if(j > 0) */ + + if (n & 1){ + aoffset1 = aoffset; + + i = m; + if (i > 0){ + do{ + ctemp01 = *(aoffset1 + 0); + ctemp02 = *(aoffset1 + 1); + + *(boffset + 0) = ctemp01; + *(boffset + 1) = ctemp02; + + aoffset1 += 2; + boffset += 2; + i --; + }while(i > 0); + } + + } /* end of if(j > 0) */ + + return 0; +}