aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp154
1 files changed, 77 insertions, 77 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp
index 76c1d053cd..6d1c1a207f 100644
--- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp
+++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_block4_u8_u8.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2022-2023 Arm Limited.
+ * Copyright (c) 2022-2024 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -32,16 +32,16 @@ void interleave_block<1, 4, VLType::SME, false>(
{
__asm__ __volatile__(
".inst 0xd503477f // SMSTART ZA\n"
- "cntb x21\n"
"mov x23, %x[width]\n"
+ "mov x21, %x[width]\n"
+ "cntb x20\n"
"incb x23\n"
- "mov x20, %x[width]\n"
- "sub x10, x21, #0x1\n"
+ "sub x10, x20, #0x1\n"
"cntw x9\n"
"sub x23, x23, #0x1\n"
- "ands x10, x20, x10\n"
- "udiv x23, x23, x21\n" // n_passes = ceildiv(width, VL<T>)
- "csel x10, x10, x21, NE\n"
+ "ands x10, x21, x10\n"
+ "udiv x23, x23, x20\n" // n_passes = ceildiv(width, VL<T>)
+ "csel x10, x10, x20, NE\n"
"lsl x22, %x[height], #0x1\n" // height * 2
"lsl x21, x9, #0x1\n"
"sub x20, x23, #0x1\n"
@@ -52,12 +52,12 @@ void interleave_block<1, 4, VLType::SME, false>(
"mov x27, #0x0\n"
"mov x26, %x[in]\n"
"lsr x20, x20, #0x1\n" // n_loops = (n_passes - 1) / 2
- "ldr x25, [x26, #0x0]\n"
- "and x24, x23, #0x1\n" // odd_tail = bool(n_passes & 0x1)
+ "and x25, x23, #0x1\n" // odd_tail = bool(n_passes & 0x1)
"lsr x10, x10, #0x2\n"
- "ldr x23, [x26, #0x8]\n"
+ "ldr x24, [x26, #0x0]\n"
"ptrue p11.s\n"
"zip1 p10.b, p9.b, p8.b\n"
+ "ldr x23, [x26, #0x8]\n"
"mov x22, %x[row_offset]\n"
"mov x21, %x[out]\n"
"whilelt p9.b, x27, %x[width]\n"
@@ -66,124 +66,124 @@ void interleave_block<1, 4, VLType::SME, false>(
"mov x12, #0x0\n"
"cbz x28, 2f\n"
"1:" // K loop: Charge: Loop
- ".inst 0x25246140 // psel p0.b, p8.b/Z, p10.b[w12]\n"
- ".inst 0xe0160320 // ld1b { za0h.b[x12] }, p0/Z, [x25, x22]\n"
+ ".inst 0x25246141 // psel p1.b, p8.b/Z, p10.b[w12]\n"
".inst 0x25646140 // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
- "ldr x25, [x26, #0x0]\n"
+ ".inst 0xe0160700 // ld1b { za0h.b[x12] }, p1/Z, [x24, x22]\n"
+ "ldr x24, [x26, #0x0]\n"
".inst 0xe01602e4 // ld1b { za0h.b[x12, #4] }, p0/Z, [x23, x22]\n"
"add x12, x12, #0x8\n"
- "cmp x12, x28, LSL #2\n"
"ldr x23, [x26, #0x8]\n"
"add x26, x26, #0x10\n"
+ "cmp x12, x28, LSL #2\n"
"blt 1b\n"
"2:" // K loop: Charge: End
- ".inst 0x25246140 // psel p0.b, p8.b/Z, p10.b[w12]\n"
- ".inst 0xe0160320 // ld1b { za0h.b[x12] }, p0/Z, [x25, x22]\n"
+ ".inst 0x25246141 // psel p1.b, p8.b/Z, p10.b[w12]\n"
".inst 0x25646140 // psel p0.b, p8.b/Z, p10.b[w12, #4]\n"
"mov x26, %x[in]\n"
+ "incb x27\n"
+ ".inst 0xe0160700 // ld1b { za0h.b[x12] }, p1/Z, [x24, x22]\n"
+ "ldr x24, [x26, #0x0]\n"
".inst 0xe01602e4 // ld1b { za0h.b[x12, #4] }, p0/Z, [x23, x22]\n"
- "ldr x25, [x26, #0x0]\n"
- "incb x22\n"
"ldr x23, [x26, #0x8]\n"
"add x26, x26, #0x10\n"
- "incb x27\n"
+ "incb x22\n"
"cbz x20, 8f\n"
"mov x20, x20\n"
"3:" // K loop: Main loop
"whilelt p8.b, x27, %x[width]\n"
- "mov x13, #0x0\n"
"mov x12, #0x0\n"
+ "mov x14, #0x0\n"
"cbz x28, 5f\n"
"4:" // K loop: Main loop: First: Loop
- ".inst 0x25356140 // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
- ".inst 0xe0162322 // ld1b { za0h.b[x13, #2] }, p0/Z, [x25, x22]\n"
- ".inst 0x25756141 // psel p1.b, p8.b/Z, p10.b[w13, #6]\n"
- "ldr x25, [x26, #0x0]\n"
- ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
- ".inst 0xe01626e6 // ld1b { za0h.b[x13, #6] }, p1/Z, [x23, x22]\n"
+ ".inst 0x25346143 // psel p3.b, p8.b/Z, p10.b[w12, #2]\n"
+ ".inst 0x25746142 // psel p2.b, p8.b/Z, p10.b[w12, #6]\n"
+ ".inst 0x25266d21 // psel p1.b, p11.b/Z, p9.b[w14]\n"
+ ".inst 0x252e6d20 // psel p0.b, p11.b/Z, p9.b[w14, #1]\n"
+ ".inst 0xe0160f02 // ld1b { za0h.b[x12, #2] }, p3/Z, [x24, x22]\n"
+ "ldr x24, [x26, #0x0]\n"
+ ".inst 0xe0160ae6 // ld1b { za0h.b[x12, #6] }, p2/Z, [x23, x22]\n"
"ldr x23, [x26, #0x8]\n"
- ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
- ".inst 0x25706d20 // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
- ".inst 0xe0a982a1 // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
- "add x12, x12, #0x2\n"
- "cmp x12, x28\n"
"add x26, x26, #0x10\n"
+ "add x12, x12, #0x8\n"
+ ".inst 0xe0bfc6a0 // st1w { za0v.s[x14] }, p1/Z, [x21, XZR, LSL #2]\n"
+ ".inst 0xe0a9c2a1 // st1w { za0v.s[x14, #1] }, p0/Z, [x21, x9, LSL #2]\n"
+ "add x14, x14, #0x2\n"
"addvl x21, x21, #2\n"
- "add x13, x13, #0x8\n"
+ "cmp x14, x28\n"
"blt 4b\n"
"5:" // K loop: Main loop: First: Tail
- ".inst 0x25356140 // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
- ".inst 0xe0162322 // ld1b { za0h.b[x13, #2] }, p0/Z, [x25, x22]\n"
+ ".inst 0x25346143 // psel p3.b, p8.b/Z, p10.b[w12, #2]\n"
+ ".inst 0x25746142 // psel p2.b, p8.b/Z, p10.b[w12, #6]\n"
+ ".inst 0x25266d21 // psel p1.b, p11.b/Z, p9.b[w14]\n"
+ ".inst 0x252e6d20 // psel p0.b, p11.b/Z, p9.b[w14, #1]\n"
"mov x26, %x[in]\n"
- "ldr x25, [x26, #0x0]\n"
- ".inst 0x25756141 // psel p1.b, p8.b/Z, p10.b[w13, #6]\n"
- ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
- ".inst 0xe01626e6 // ld1b { za0h.b[x13, #6] }, p1/Z, [x23, x22]\n"
- "ldr x23, [x26, #0x8]\n"
- ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
- ".inst 0x25706d20 // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
"whilelt p9.b, x27, %x[width]\n"
+ ".inst 0xe0160f02 // ld1b { za0h.b[x12, #2] }, p3/Z, [x24, x22]\n"
+ "ldr x24, [x26, #0x0]\n"
"incb x27\n"
+ "mov x13, #0x0\n"
+ ".inst 0xe0160ae6 // ld1b { za0h.b[x12, #6] }, p2/Z, [x23, x22]\n"
+ "ldr x23, [x26, #0x8]\n"
"add x26, x26, #0x10\n"
- ".inst 0xe0a982a1 // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
- "addvl x21, x21, #2\n"
"incb x22\n"
+ ".inst 0xe0bfc6a0 // st1w { za0v.s[x14] }, p1/Z, [x21, XZR, LSL #2]\n"
"whilelt p8.b, x27, %x[width]\n"
- "mov x13, #0x0\n"
"mov x12, #0x0\n"
+ ".inst 0xe0a9c2a1 // st1w { za0v.s[x14, #1] }, p0/Z, [x21, x9, LSL #2]\n"
+ "addvl x21, x21, #2\n"
"cbz x28, 7f\n"
"6:" // K loop: Main loop: Second: Loop
- ".inst 0x25256140 // psel p0.b, p8.b/Z, p10.b[w13]\n"
- ".inst 0xe0162320 // ld1b { za0h.b[x13] }, p0/Z, [x25, x22]\n"
- ".inst 0x25656141 // psel p1.b, p8.b/Z, p10.b[w13, #4]\n"
- "ldr x25, [x26, #0x0]\n"
- ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
- ".inst 0xe01626e4 // ld1b { za0h.b[x13, #4] }, p1/Z, [x23, x22]\n"
+ ".inst 0x25256143 // psel p3.b, p8.b/Z, p10.b[w13]\n"
+ ".inst 0x25656142 // psel p2.b, p8.b/Z, p10.b[w13, #4]\n"
+ ".inst 0x25246d21 // psel p1.b, p11.b/Z, p9.b[w12]\n"
+ ".inst 0x252c6d20 // psel p0.b, p11.b/Z, p9.b[w12, #1]\n"
+ ".inst 0xe0162f00 // ld1b { za0h.b[x13] }, p3/Z, [x24, x22]\n"
+ "ldr x24, [x26, #0x0]\n"
+ ".inst 0xe0162ae4 // ld1b { za0h.b[x13, #4] }, p2/Z, [x23, x22]\n"
"ldr x23, [x26, #0x8]\n"
- ".inst 0xe0bf82a8 // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
- ".inst 0x25706d20 // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
+ "add x26, x26, #0x10\n"
+ "add x13, x13, #0x8\n"
+ ".inst 0xe0bf86a8 // st1w { za2v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
".inst 0xe0a982a9 // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
"add x12, x12, #0x2\n"
- "cmp x12, x28\n"
- "add x26, x26, #0x10\n"
"addvl x21, x21, #2\n"
- "add x13, x13, #0x8\n"
+ "cmp x12, x28\n"
"blt 6b\n"
"7:" // K loop: Main loop: Second: Tail
- ".inst 0x25256140 // psel p0.b, p8.b/Z, p10.b[w13]\n"
- ".inst 0xe0162320 // ld1b { za0h.b[x13] }, p0/Z, [x25, x22]\n"
+ ".inst 0x25256143 // psel p3.b, p8.b/Z, p10.b[w13]\n"
+ ".inst 0x25656142 // psel p2.b, p8.b/Z, p10.b[w13, #4]\n"
+ ".inst 0x25246d21 // psel p1.b, p11.b/Z, p9.b[w12]\n"
+ ".inst 0x252c6d20 // psel p0.b, p11.b/Z, p9.b[w12, #1]\n"
"mov x26, %x[in]\n"
- "ldr x25, [x26, #0x0]\n"
- ".inst 0x25656141 // psel p1.b, p8.b/Z, p10.b[w13, #4]\n"
- ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
- ".inst 0xe01626e4 // ld1b { za0h.b[x13, #4] }, p1/Z, [x23, x22]\n"
- "ldr x23, [x26, #0x8]\n"
- ".inst 0xe0bf82a8 // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
- ".inst 0x25706d20 // psel p0.s, p11.s/Z, p9.s[w12, #1]\n"
"whilelt p9.b, x27, %x[width]\n"
+ ".inst 0xe0162f00 // ld1b { za0h.b[x13] }, p3/Z, [x24, x22]\n"
+ "ldr x24, [x26, #0x0]\n"
"subs x20, x20, #0x1\n"
+ "incb x27\n"
+ ".inst 0xe0162ae4 // ld1b { za0h.b[x13, #4] }, p2/Z, [x23, x22]\n"
+ "ldr x23, [x26, #0x8]\n"
"add x26, x26, #0x10\n"
+ "incb x22\n"
+ ".inst 0xe0bf86a8 // st1w { za2v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
".inst 0xe0a982a9 // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x9, LSL #2]\n"
"addvl x21, x21, #2\n"
- "incb x27\n"
- "incb x22\n"
"bgt 3b\n"
"8:" // K loop: Tails
- "cbnz x24, 11f\n"
+ "cbnz x25, 11f\n"
"mov x26, %x[in]\n"
"whilelt p8.b, x27, %x[width]\n"
"mov x13, #0x0\n"
"mov x12, #0x0\n"
"9:" // K loop: Tails: Even: First
- ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
- ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
- "ldr x20, [x26, #0x0]\n"
- "add x12, x12, #0x1\n"
+ ".inst 0x25306d21 // psel p1.s, p11.s/Z, p9.s[w12]\n"
".inst 0x25356140 // psel p0.b, p8.b/Z, p10.b[w13, #2]\n"
+ ".inst 0xe0bf86a0 // st1w { za0v.s[x12] }, p1/Z, [x21, XZR, LSL #2]\n"
+ "add x12, x12, #0x1\n"
+ "addvl x21, x21, #1\n"
+ "ldr x20, [x26, #0x0]\n"
"cmp x12, x9\n"
- ".inst 0xe0162282 // ld1b { za0h.b[x13, #2] }, p0/Z, [x20, x22]\n"
"add x26, x26, #0x8\n"
- "addvl x21, x21, #1\n"
+ ".inst 0xe0162282 // ld1b { za0h.b[x13, #2] }, p0/Z, [x20, x22]\n"
"add x13, x13, #0x4\n"
"blt 9b\n"
"whilelt p9.b, x27, %x[width]\n"
@@ -192,11 +192,11 @@ void interleave_block<1, 4, VLType::SME, false>(
"mov x12, #0x0\n"
"10:" // K loop: Tails: Even: Second
".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
+ "add x20, x20, #0x4\n"
".inst 0xe0bf82a8 // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
"add x12, x12, #0x1\n"
- "cmp x12, x10\n"
"addvl x21, x21, #1\n"
- "add x20, x20, #0x4\n"
+ "cmp x12, x10\n"
"blt 10b\n"
"whilelt p8.b, x27, %x[width]\n"
"b 13f\n"
@@ -206,15 +206,15 @@ void interleave_block<1, 4, VLType::SME, false>(
".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n"
"add x12, x12, #0x1\n"
- "cmp x12, x10\n"
"addvl x21, x21, #1\n"
+ "cmp x12, x10\n"
"blt 12b\n"
"13:" // K loop: End
"mov %x[out], x21\n"
".inst 0xd503467f // SMSTOP\n"
: [out] "+&r" (out)
: [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
- : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x13", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
+ : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x13", "x14", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
);
}
generated by cgit v1.2.1 (git 2.23.0) at 2024-07-07 17:42:50 +0000