aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_u8_nhwc_max_generic_depthfirst/generic.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/kernels/sme_u8_nhwc_max_generic_depthfirst/generic.cpp')
-rw-r--r--src/core/NEON/kernels/arm_conv/pooling/kernels/sme_u8_nhwc_max_generic_depthfirst/generic.cpp151
1 files changed, 76 insertions, 75 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_u8_nhwc_max_generic_depthfirst/generic.cpp b/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_u8_nhwc_max_generic_depthfirst/generic.cpp
index 9a13deafda..06f13e8111 100644
--- a/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_u8_nhwc_max_generic_depthfirst/generic.cpp
+++ b/src/core/NEON/kernels/arm_conv/pooling/kernels/sme_u8_nhwc_max_generic_depthfirst/generic.cpp
@@ -22,9 +22,10 @@
* SOFTWARE.
*/
-#if defined(ARM_COMPUTE_ENABLE_SME)
-
#include <cstdint>
+#include <cstddef>
+
+#if defined(ARM_COMPUTE_ENABLE_SME)
namespace arm_conv {
namespace pooling {
@@ -54,68 +55,68 @@ void sme_u8_nhwc_max_generic_depthfirst_impl(
"lsr x25, %x[n_valid_cells], #0x2\n"
"mov z4.b, #0x0\n"
"mov z3.b, #0x0\n"
- "mov x20, %x[inptrs]\n"
+ "mov x24, %x[inptrs]\n"
"mov z2.b, #0x0\n"
"mov z1.b, #0x0\n"
"cbz x25, 4f\n"
- "ldp x24, x23, [x20, #0x0]\n"
+ "ldp x23, x22, [x24, #0x0]\n"
"subs x25, x25, #0x1\n"
- "ld1b { z0.b }, p4/Z, [x24, x9]\n"
- "ldp x22, x21, [x20, #0x10]\n"
- "add x20, x20, #0x20\n"
- "ld1b { z31.b }, p4/Z, [x23, x9]\n"
- "ld1b { z23.b }, p4/Z, [x22, x9]\n"
- "ld1b { z30.b }, p4/Z, [x21, x9]\n"
- "ld1b { z18.b }, p3/Z, [x24, x28]\n"
- "ld1b { z29.b }, p3/Z, [x23, x28]\n"
- "ld1b { z22.b }, p3/Z, [x22, x28]\n"
- "ld1b { z28.b }, p3/Z, [x21, x28]\n"
- "ld1b { z17.b }, p2/Z, [x24, x27]\n"
- "ld1b { z27.b }, p2/Z, [x23, x27]\n"
- "ld1b { z21.b }, p2/Z, [x22, x27]\n"
- "ld1b { z26.b }, p2/Z, [x21, x27]\n"
- "ld1b { z16.b }, p1/Z, [x24, x26]\n"
- "ld1b { z25.b }, p1/Z, [x23, x26]\n"
- "ld1b { z20.b }, p1/Z, [x22, x26]\n"
- "ld1b { z24.b }, p1/Z, [x21, x26]\n"
+ "ld1b { z0.b }, p4/Z, [x23, x9]\n"
+ "ldp x21, x20, [x24, #0x10]\n"
+ "add x24, x24, #0x20\n"
+ "ld1b { z31.b }, p4/Z, [x22, x9]\n"
+ "ld1b { z23.b }, p4/Z, [x21, x9]\n"
+ "ld1b { z30.b }, p4/Z, [x20, x9]\n"
+ "ld1b { z18.b }, p3/Z, [x23, x28]\n"
+ "ld1b { z29.b }, p3/Z, [x22, x28]\n"
+ "ld1b { z22.b }, p3/Z, [x21, x28]\n"
+ "ld1b { z28.b }, p3/Z, [x20, x28]\n"
+ "ld1b { z17.b }, p2/Z, [x23, x27]\n"
+ "ld1b { z27.b }, p2/Z, [x22, x27]\n"
+ "ld1b { z21.b }, p2/Z, [x21, x27]\n"
+ "ld1b { z26.b }, p2/Z, [x20, x27]\n"
+ "ld1b { z16.b }, p1/Z, [x23, x26]\n"
+ "ld1b { z25.b }, p1/Z, [x22, x26]\n"
+ "ld1b { z20.b }, p1/Z, [x21, x26]\n"
+ "ld1b { z24.b }, p1/Z, [x20, x26]\n"
"beq 3f\n"
"2:" // 4-vectors of channels: 4 inputs loop
"movprfx z19, z0\n umax z19.b, p0/M, z19.b, z31.b\n"
"umax z23.b, p0/M, z23.b, z30.b\n"
- "ldp x24, x23, [x20, #0x0]\n"
+ "ldp x23, x22, [x24, #0x0]\n"
"subs x25, x25, #0x1\n"
"umax z18.b, p0/M, z18.b, z29.b\n"
"umax z22.b, p0/M, z22.b, z28.b\n"
- "ldp x22, x21, [x20, #0x10]\n"
- "add x20, x20, #0x20\n"
+ "ldp x21, x20, [x24, #0x10]\n"
+ "add x24, x24, #0x20\n"
"umax z17.b, p0/M, z17.b, z27.b\n"
"umax z21.b, p0/M, z21.b, z26.b\n"
- "ld1b { z0.b }, p4/Z, [x24, x9]\n"
+ "ld1b { z0.b }, p4/Z, [x23, x9]\n"
"umax z16.b, p0/M, z16.b, z25.b\n"
"umax z20.b, p0/M, z20.b, z24.b\n"
- "ld1b { z31.b }, p4/Z, [x23, x9]\n"
+ "ld1b { z31.b }, p4/Z, [x22, x9]\n"
"umax z19.b, p0/M, z19.b, z23.b\n"
"umax z18.b, p0/M, z18.b, z22.b\n"
- "ld1b { z23.b }, p4/Z, [x22, x9]\n"
+ "ld1b { z23.b }, p4/Z, [x21, x9]\n"
"umax z17.b, p0/M, z17.b, z21.b\n"
"umax z16.b, p0/M, z16.b, z20.b\n"
- "ld1b { z30.b }, p4/Z, [x21, x9]\n"
+ "ld1b { z30.b }, p4/Z, [x20, x9]\n"
"umax z4.b, p0/M, z4.b, z19.b\n"
"umax z3.b, p0/M, z3.b, z18.b\n"
- "ld1b { z18.b }, p3/Z, [x24, x28]\n"
+ "ld1b { z18.b }, p3/Z, [x23, x28]\n"
"umax z2.b, p0/M, z2.b, z17.b\n"
"umax z1.b, p0/M, z1.b, z16.b\n"
- "ld1b { z29.b }, p3/Z, [x23, x28]\n"
- "ld1b { z22.b }, p3/Z, [x22, x28]\n"
- "ld1b { z28.b }, p3/Z, [x21, x28]\n"
- "ld1b { z17.b }, p2/Z, [x24, x27]\n"
- "ld1b { z27.b }, p2/Z, [x23, x27]\n"
- "ld1b { z21.b }, p2/Z, [x22, x27]\n"
- "ld1b { z26.b }, p2/Z, [x21, x27]\n"
- "ld1b { z16.b }, p1/Z, [x24, x26]\n"
- "ld1b { z25.b }, p1/Z, [x23, x26]\n"
- "ld1b { z20.b }, p1/Z, [x22, x26]\n"
- "ld1b { z24.b }, p1/Z, [x21, x26]\n"
+ "ld1b { z29.b }, p3/Z, [x22, x28]\n"
+ "ld1b { z22.b }, p3/Z, [x21, x28]\n"
+ "ld1b { z28.b }, p3/Z, [x20, x28]\n"
+ "ld1b { z17.b }, p2/Z, [x23, x27]\n"
+ "ld1b { z27.b }, p2/Z, [x22, x27]\n"
+ "ld1b { z21.b }, p2/Z, [x21, x27]\n"
+ "ld1b { z26.b }, p2/Z, [x20, x27]\n"
+ "ld1b { z16.b }, p1/Z, [x23, x26]\n"
+ "ld1b { z25.b }, p1/Z, [x22, x26]\n"
+ "ld1b { z20.b }, p1/Z, [x21, x26]\n"
+ "ld1b { z24.b }, p1/Z, [x20, x26]\n"
"bgt 2b\n"
"3:" // 4-vectors of channels: 4 inputs tail
"movprfx z19, z0\n umax z19.b, p0/M, z19.b, z31.b\n"
@@ -138,15 +139,15 @@ void sme_u8_nhwc_max_generic_depthfirst_impl(
"ands x21, %x[n_valid_cells], #0x3\n"
"beq 6f\n"
"5:" // 4-vectors of channels: Single input loop
- "ldr x24, [x20], #0x8\n"
- "ld1b { z0.b }, p4/Z, [x24, x9]\n"
+ "ldr x20, [x24], #0x8\n"
+ "ld1b { z16.b }, p4/Z, [x20, x9]\n"
"subs x21, x21, #0x1\n"
- "umax z4.b, p0/M, z4.b, z0.b\n"
- "ld1b { z18.b }, p3/Z, [x24, x28]\n"
- "umax z3.b, p0/M, z3.b, z18.b\n"
- "ld1b { z17.b }, p2/Z, [x24, x27]\n"
- "umax z2.b, p0/M, z2.b, z17.b\n"
- "ld1b { z16.b }, p1/Z, [x24, x26]\n"
+ "umax z4.b, p0/M, z4.b, z16.b\n"
+ "ld1b { z16.b }, p3/Z, [x20, x28]\n"
+ "umax z3.b, p0/M, z3.b, z16.b\n"
+ "ld1b { z16.b }, p2/Z, [x20, x27]\n"
+ "umax z2.b, p0/M, z2.b, z16.b\n"
+ "ld1b { z16.b }, p1/Z, [x20, x26]\n"
"umax z1.b, p0/M, z1.b, z16.b\n"
"bgt 5b\n"
"6:" // 4-vectors of channels: Single input loop: End
@@ -166,44 +167,44 @@ void sme_u8_nhwc_max_generic_depthfirst_impl(
"8:" // Single vector of channels: Loop
"lsr x25, %x[n_valid_cells], #0x2\n"
"mov z4.b, #0x0\n"
- "mov x20, %x[inptrs]\n"
+ "mov x24, %x[inptrs]\n"
"cbz x25, 11f\n"
- "ldp x24, x23, [x20, #0x0]\n"
+ "ldp x20, x22, [x24, #0x0]\n"
"subs x25, x25, #0x1\n"
- "ld1b { z0.b }, p4/Z, [x24, x9]\n"
- "ldp x22, x21, [x20, #0x10]\n"
- "add x20, x20, #0x20\n"
- "ld1b { z31.b }, p4/Z, [x23, x9]\n"
- "ld1b { z23.b }, p4/Z, [x22, x9]\n"
- "ld1b { z30.b }, p4/Z, [x21, x9]\n"
+ "ld1b { z0.b }, p4/Z, [x20, x9]\n"
+ "ldp x21, x20, [x24, #0x10]\n"
+ "add x24, x24, #0x20\n"
+ "ld1b { z31.b }, p4/Z, [x22, x9]\n"
+ "ld1b { z23.b }, p4/Z, [x21, x9]\n"
+ "ld1b { z30.b }, p4/Z, [x20, x9]\n"
"beq 10f\n"
"9:" // Single vector of channels: Loop: 4 inputs loop
- "movprfx z19, z0\n umax z19.b, p0/M, z19.b, z31.b\n"
- "umax z23.b, p0/M, z23.b, z30.b\n"
- "ldp x24, x23, [x20, #0x0]\n"
+ "movprfx z16, z0\n umax z16.b, p0/M, z16.b, z31.b\n"
+ "movprfx z17, z23\n umax z17.b, p0/M, z17.b, z30.b\n"
+ "ldp x23, x22, [x24, #0x0]\n"
"subs x25, x25, #0x1\n"
- "umax z19.b, p0/M, z19.b, z23.b\n"
- "ldp x22, x21, [x20, #0x10]\n"
- "umax z4.b, p0/M, z4.b, z19.b\n"
- "add x20, x20, #0x20\n"
- "ld1b { z0.b }, p4/Z, [x24, x9]\n"
- "ld1b { z31.b }, p4/Z, [x23, x9]\n"
- "ld1b { z23.b }, p4/Z, [x22, x9]\n"
- "ld1b { z30.b }, p4/Z, [x21, x9]\n"
+ "umax z16.b, p0/M, z16.b, z17.b\n"
+ "ldp x21, x20, [x24, #0x10]\n"
+ "umax z4.b, p0/M, z4.b, z16.b\n"
+ "add x24, x24, #0x20\n"
+ "ld1b { z0.b }, p4/Z, [x23, x9]\n"
+ "ld1b { z31.b }, p4/Z, [x22, x9]\n"
+ "ld1b { z23.b }, p4/Z, [x21, x9]\n"
+ "ld1b { z30.b }, p4/Z, [x20, x9]\n"
"bgt 9b\n"
"10:" // Single vector of channels: Loop: 4 inputs tail
- "movprfx z19, z0\n umax z19.b, p0/M, z19.b, z31.b\n"
- "umax z23.b, p0/M, z23.b, z30.b\n"
- "umax z19.b, p0/M, z19.b, z23.b\n"
- "umax z4.b, p0/M, z4.b, z19.b\n"
+ "movprfx z16, z0\n umax z16.b, p0/M, z16.b, z31.b\n"
+ "movprfx z17, z23\n umax z17.b, p0/M, z17.b, z30.b\n"
+ "umax z16.b, p0/M, z16.b, z17.b\n"
+ "umax z4.b, p0/M, z4.b, z16.b\n"
"11:" // Single vector of channels: Loop: After loop
"ands x21, %x[n_valid_cells], #0x3\n"
"beq 13f\n"
"12:" // Single vector of channels: Loop: Single input loop
- "ldr x24, [x20], #0x8\n"
- "ld1b { z0.b }, p4/Z, [x24, x9]\n"
+ "ldr x20, [x24], #0x8\n"
+ "ld1b { z16.b }, p4/Z, [x20, x9]\n"
"subs x21, x21, #0x1\n"
- "umax z4.b, p0/M, z4.b, z0.b\n"
+ "umax z4.b, p0/M, z4.b, z16.b\n"
"bgt 12b\n"
"13:" // Single vector of channels: Loop: Single input loop: End
"st1b { z4.b }, p4, [%x[outptr], x9]\n"
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-09 21:25:01 +0000