Add additional test cases

Author: NickGuy-Arm

llvm/test/CodeGen/AArch64/sve-fixed-length-partial-reduce.ll

@@ -434,6 +434,127 @@ define <4 x i32> @four_way_i8_i32_vl128_usdot(ptr %accptr, ptr %uptr, ptr %sptr)
   ret <4 x i32> %partial.reduce
 }
 
+define <4 x i32> @four_way_i8_i32_vl128_sudot(ptr %accptr, ptr %uptr, ptr %sptr) {
+; COMMON-LABEL: four_way_i8_i32_vl128_sudot:
+; COMMON:       // %bb.0:
+; COMMON-NEXT:    ldr q0, [x0]
+; COMMON-NEXT:    ldr q1, [x1]
+; COMMON-NEXT:    ldr q2, [x2]
+; COMMON-NEXT:    usdot v0.4s, v2.16b, v1.16b
+; COMMON-NEXT:    ret
+;
+; SME-LABEL: four_way_i8_i32_vl128_sudot:
+; SME:       // %bb.0:
+; SME-NEXT:    ldr q0, [x0]
+; SME-NEXT:    ldr q1, [x1]
+; SME-NEXT:    ldr q2, [x2]
+; SME-NEXT:    usdot z0.s, z2.b, z1.b
+; SME-NEXT:    // kill: def $q0 killed $q0 killed $z0
+; SME-NEXT:    ret
+  %acc = load <4 x i32>, ptr %accptr
+  %u = load <16 x i8>, ptr %uptr
+  %s = load <16 x i8>, ptr %sptr
+  %u.wide = sext <16 x i8> %u to <16 x i32>
+  %s.wide = zext <16 x i8> %s to <16 x i32>
+  %mult = mul nuw nsw <16 x i32> %s.wide, %u.wide
+  %partial.reduce = tail call <4 x i32> @llvm.experimental.vector.partial.reduce.add(<4 x i32> %acc, <16 x i32> %mult)
+  ret <4 x i32> %partial.reduce
+}
+
+define <2 x i64> @four_way_i8_i64_vl128_usdot(ptr %accptr, ptr %uptr, ptr %sptr) {
+; NEON-LABEL: four_way_i8_i64_vl128_usdot:
+; NEON:       // %bb.0:
+; NEON-NEXT:    movi v0.2d, #0000000000000000
+; NEON-NEXT:    ldr q1, [x1]
+; NEON-NEXT:    ldr q2, [x2]
+; NEON-NEXT:    usdot v0.4s, v1.16b, v2.16b
+; NEON-NEXT:    ldr q1, [x0]
+; NEON-NEXT:    saddw v1.2d, v1.2d, v0.2s
+; NEON-NEXT:    saddw2 v0.2d, v1.2d, v0.4s
+; NEON-NEXT:    ret
+;
+; SVE-LABEL: four_way_i8_i64_vl128_usdot:
+; SVE:       // %bb.0:
+; SVE-NEXT:    movi v0.2d, #0000000000000000
+; SVE-NEXT:    ldr q1, [x1]
+; SVE-NEXT:    ldr q2, [x2]
+; SVE-NEXT:    usdot z0.s, z1.b, z2.b
+; SVE-NEXT:    ldr q2, [x0]
+; SVE-NEXT:    sunpklo z1.d, z0.s
+; SVE-NEXT:    sunpkhi z0.d, z0.s
+; SVE-NEXT:    add z1.d, z2.d, z1.d
+; SVE-NEXT:    add z0.d, z1.d, z0.d
+; SVE-NEXT:    // kill: def $q0 killed $q0 killed $z0
+; SVE-NEXT:    ret
+;
+; SME-LABEL: four_way_i8_i64_vl128_usdot:
+; SME:       // %bb.0:
+; SME-NEXT:    mov z0.s, #0 // =0x0
+; SME-NEXT:    ldr q1, [x1]
+; SME-NEXT:    ldr q2, [x2]
+; SME-NEXT:    usdot z0.s, z1.b, z2.b
+; SME-NEXT:    ldr q1, [x0]
+; SME-NEXT:    saddwb z1.d, z1.d, z0.s
+; SME-NEXT:    saddwt z0.d, z1.d, z0.s
+; SME-NEXT:    // kill: def $q0 killed $q0 killed $z0
+; SME-NEXT:    ret
+  %acc = load <2 x i64>, ptr %accptr
+  %u = load <16 x i8>, ptr %uptr
+  %s = load <16 x i8>, ptr %sptr
+  %u.wide = zext <16 x i8> %u to <16 x i64>
+  %s.wide = sext <16 x i8> %s to <16 x i64>
+  %mult = mul nuw nsw <16 x i64> %s.wide, %u.wide
+  %partial.reduce = tail call <2 x i64> @llvm.experimental.vector.partial.reduce.add(<2 x i64> %acc, <16 x i64> %mult)
+  ret <2 x i64> %partial.reduce
+}
+
+define <2 x i64> @four_way_i16_i64_vl128_usdot(ptr %accptr, ptr %uptr, ptr %sptr) {
+; COMMON-LABEL: four_way_i16_i64_vl128_usdot:
+; COMMON:       // %bb.0:
+; COMMON-NEXT:    ldr q1, [x1]
+; COMMON-NEXT:    ldr q2, [x2]
+; COMMON-NEXT:    ldr q0, [x0]
+; COMMON-NEXT:    ushll v3.4s, v1.4h, #0
+; COMMON-NEXT:    sshll v4.4s, v2.4h, #0
+; COMMON-NEXT:    ushll2 v1.4s, v1.8h, #0
+; COMMON-NEXT:    sshll2 v2.4s, v2.8h, #0
+; COMMON-NEXT:    smlal v0.2d, v4.2s, v3.2s
+; COMMON-NEXT:    smlal2 v0.2d, v4.4s, v3.4s
+; COMMON-NEXT:    smlal v0.2d, v2.2s, v1.2s
+; COMMON-NEXT:    smlal2 v0.2d, v2.4s, v1.4s
+; COMMON-NEXT:    ret
+;
+; SME-LABEL: four_way_i16_i64_vl128_usdot:
+; SME:       // %bb.0:
+; SME-NEXT:    ptrue p0.d, vl2
+; SME-NEXT:    ldr q2, [x0]
+; SME-NEXT:    mov x8, #2 // =0x2
+; SME-NEXT:    ld1h { z0.d }, p0/z, [x1]
+; SME-NEXT:    ld1sh { z1.d }, p0/z, [x2]
+; SME-NEXT:    mad z0.d, p0/m, z1.d, z2.d
+; SME-NEXT:    ld1h { z1.d }, p0/z, [x1, x8, lsl #1]
+; SME-NEXT:    ld1sh { z2.d }, p0/z, [x2, x8, lsl #1]
+; SME-NEXT:    mov x8, #4 // =0x4
+; SME-NEXT:    mla z0.d, p0/m, z2.d, z1.d
+; SME-NEXT:    ld1h { z1.d }, p0/z, [x1, x8, lsl #1]
+; SME-NEXT:    ld1sh { z2.d }, p0/z, [x2, x8, lsl #1]
+; SME-NEXT:    mov x8, #6 // =0x6
+; SME-NEXT:    mla z0.d, p0/m, z2.d, z1.d
+; SME-NEXT:    ld1h { z1.d }, p0/z, [x1, x8, lsl #1]
+; SME-NEXT:    ld1sh { z2.d }, p0/z, [x2, x8, lsl #1]
+; SME-NEXT:    mla z0.d, p0/m, z2.d, z1.d
+; SME-NEXT:    // kill: def $q0 killed $q0 killed $z0
+; SME-NEXT:    ret
+  %acc = load <2 x i64>, ptr %accptr
+  %u = load <8 x i16>, ptr %uptr
+  %s = load <8 x i16>, ptr %sptr
+  %u.wide = zext <8 x i16> %u to <8 x i64>
+  %s.wide = sext <8 x i16> %s to <8 x i64>
+  %mult = mul nuw nsw <8 x i64> %s.wide, %u.wide
+  %partial.reduce = tail call <2 x i64> @llvm.experimental.vector.partial.reduce.add(<2 x i64> %acc, <8 x i64> %mult)
+  ret <2 x i64> %partial.reduce
+}
+
 define <8 x i32> @four_way_i8_i32_vl128_double_width(ptr %accptr, ptr %uptr, ptr %sptr) {
 ;
 ; COMMON-LABEL: four_way_i8_i32_vl128_double_width: