ARM32 vector lowering: fabs, scalarize remaining arithmetic operations.

tests_lit/assembler/arm32/rem-vec.ll

+; Show that we know how to translate vector urem, srem and frem.
+
+; NOTE: We use -O2 to get rid of memory stores.
+
+; REQUIRES: allow_dump
+
+; Compile using standalone assembler.
+; RUN: %p2i --filetype=asm -i %s --target=arm32 --args -O2 -mattr=hwdiv-arm \
+; RUN:   | FileCheck %s --check-prefix=ASM
+
+define internal <4 x i32> @Urem4i32(<4 x i32> %a, <4 x i32> %b) {
+; ASM-LABEL:Urem4i32:
+
+entry:
+; ASM-NEXT:.LUrem4i32$entry:
+
+  %v = urem <4 x i32> %a, %b
+
+; ASM-LABEL:.LUrem4i32$local$__0:
+; ASM-NEXT:     udiv    r2, r0, r1
+; ASM-NEXT:     mls     r2, r2, r1, r0
+; ASM-NEXT:     vmov.32 d4[0], r2
+
+; ASM-LABEL:.LUrem4i32$local$__1:
+; ASM-NEXT:     udiv    r2, r0, r1
+; ASM-NEXT:     mls     r2, r2, r1, r0
+; ASM-NEXT:     vmov.32 d4[1], r2
+
+; ASM-LABEL:.LUrem4i32$local$__2:
+; ASM-NEXT:     udiv    r2, r0, r1
+; ASM-NEXT:     mls     r2, r2, r1, r0
+; ASM-NEXT:     vmov.32 d5[0], r2
+
+; ASM-LABEL:.LUrem4i32$local$__3:
+; ASM-NEXT:     udiv    r2, r0, r1
+; ASM-NEXT:     mls     r2, r2, r1, r0
+; ASM-NEXT:     vmov.32 d5[1], r2
+
+  ret <4 x i32> %v
+}
+
+define internal <4 x i32> @Srem4i32(<4 x i32> %a, <4 x i32> %b) {
+; ASM-LABEL:Srem4i32:
+
+entry:
+; ASM-NEXT:.LSrem4i32$entry:
+
+  %v = srem <4 x i32> %a, %b
+
+; ASM-LABEL:.LSrem4i32$local$__0:
+; ASM-NEXT:     sdiv    r2, r0, r1
+; ASM-NEXT:     mls     r2, r2, r1, r0
+; ASM-NEXT:     vmov.32 d4[0], r2
+
+; ASM-LABEL:.LSrem4i32$local$__1:
+; ASM-NEXT:     sdiv    r2, r0, r1
+; ASM-NEXT:     mls     r2, r2, r1, r0
+; ASM-NEXT:     vmov.32 d4[1], r2
+
+; ASM-LABEL:.LSrem4i32$local$__2:
+; ASM-NEXT:     sdiv    r2, r0, r1
+; ASM-NEXT:     mls     r2, r2, r1, r0
+; ASM-NEXT:     vmov.32 d5[0], r2
+
+; ASM-LABEL:.LSrem4i32$local$__3:
+; ASM-NEXT:     sdiv    r2, r0, r1
+; ASM-NEXT:     mls     r2, r2, r1, r0
+; ASM-NEXT:     vmov.32 d5[1], r2
+
+  ret <4 x i32> %v
+}
+
+define internal <4 x float> @Frem4float(<4 x float> %a, <4 x float> %b) {
+; ASM-LABEL:Frem4float:
+
+entry:
+; ASM-NEXT:.LFrem4float$entry:
+
+  %v = frem <4 x float> %a, %b
+
+; ASM:          vmov.f32        s0, s16
+; ASM-NEXT:     vmov.f32        s1, s20
+; ASM-NEXT:     movw    r0, #:lower16:fmodf
+; ASM-NEXT:     movt    r0, #:upper16:fmodf
+; ASM-NEXT:     blx     r0
+
+; ASM:          vmov.f32        s0, s17
+; ASM-NEXT:     vmov.f32        s1, s21
+; ASM-NEXT:     movw    r0, #:lower16:fmodf
+; ASM-NEXT:     movt    r0, #:upper16:fmodf
+; ASM-NEXT:     blx     r0
+
+; ASM:          vmov.f32        s0, s18
+; ASM-NEXT:     vmov.f32        s1, s22
+; ASM-NEXT:     movw    r0, #:lower16:fmodf
+; ASM-NEXT:     movt    r0, #:upper16:fmodf
+; ASM-NEXT:     blx     r0
+
+; ASM:          vmov.f32        s16, s19
+; ASM-NEXT:     vmov.f32        s20, s23
+; ASM-NEXT:     movw    r0, #:lower16:fmodf
+; ASM-NEXT:     movt    r0, #:upper16:fmodf
+; ASM:  blx     r0
+
+  ret <4 x float> %v
+}
+

[Jim Stichnoth, 2016/02/17 04:15:44]

trailing blank line