[llvm] [AMDGPU] Eliminate unnecessary packing in wider f16 vectors for sdwa/opsel-able instruction (PR #137137)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,2758 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX8 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX906 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX908 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx942 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX942 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1030 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11 %s
+
+declare <1 x half> @llvm.sin.v1f16(<1 x half>)
+declare <1 x half> @llvm.cos.v1f16(<1 x half>)
+declare <1 x half> @llvm.log.v1f16(<1 x half>)
+declare <1 x half> @llvm.log2.v1f16(<1 x half>)
+declare <1 x half> @llvm.log10.v1f16(<1 x half>)
+declare <1 x half> @llvm.exp.v1f16(<1 x half>)
+declare <1 x half> @llvm.exp2.v1f16(<1 x half>)
+declare <1 x half> @llvm.exp10.v1f16(<1 x half>)
+declare <1 x half> @llvm.sqrt.v1f16(<1 x half>)
+
+declare <2 x half> @llvm.sin.v2f16(<2 x half>)
+declare <2 x half> @llvm.cos.v2f16(<2 x half>)
+declare <2 x half> @llvm.log.v2f16(<2 x half>)
+declare <2 x half> @llvm.log2.v2f16(<2 x half>)
+declare <2 x half> @llvm.log10.v2f16(<2 x half>)
+declare <2 x half> @llvm.exp.v2f16(<2 x half>)
+declare <2 x half> @llvm.exp2.v2f16(<2 x half>)
+declare <2 x half> @llvm.exp10.v2f16(<2 x half>)
+declare <2 x half> @llvm.sqrt.v2f16(<2 x half>)
+
+declare <4 x half> @llvm.sin.v4f16(<4 x half>)
+declare <4 x half> @llvm.cos.v4f16(<4 x half>)
+declare <4 x half> @llvm.log.v4f16(<4 x half>)
+declare <4 x half> @llvm.log2.v4f16(<4 x half>)
+declare <4 x half> @llvm.log10.v4f16(<4 x half>)
+declare <4 x half> @llvm.exp.v4f16(<4 x half>)
+declare <4 x half> @llvm.exp2.v4f16(<4 x half>)
+declare <4 x half> @llvm.exp10.v4f16(<4 x half>)
+declare <4 x half> @llvm.sqrt.v4f16(<4 x half>)
+
+declare <5 x half> @llvm.sin.v5f16(<5 x half>)
+declare <5 x half> @llvm.cos.v5f16(<5 x half>)
+declare <5 x half> @llvm.log.v5f16(<5 x half>)
+declare <5 x half> @llvm.log2.v5f16(<5 x half>)
+declare <5 x half> @llvm.log10.v5f16(<5 x half>)
+declare <5 x half> @llvm.exp.v5f16(<5 x half>)
+declare <5 x half> @llvm.exp2.v5f16(<5 x half>)
+declare <5 x half> @llvm.exp10.v5f16(<5 x half>)
+declare <5 x half> @llvm.sqrt.v5f16(<5 x half>)
+
+
+define <1 x half> @sin_v1f16(<1 x half> %a) {
+; GFX8-LABEL: sin_v1f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_mul_f16_e32 v0, 0.15915494, v0
+; GFX8-NEXT:    v_fract_f16_e32 v0, v0
+; GFX8-NEXT:    v_sin_f16_e32 v0, v0
+; GFX8-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: sin_v1f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mul_f16_e32 v0, 0.15915494, v0
+; GFX9-NEXT:    v_sin_f16_e32 v0, v0
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: sin_v1f16:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    v_mul_f16_e32 v0, 0.15915494, v0
+; GFX10-NEXT:    v_sin_f16_e32 v0, v0
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: sin_v1f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    v_mul_f16_e32 v0, 0.15915494, v0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_sin_f16_e32 v0, v0
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+	%res = call <1 x half> @llvm.sin.v1f16(<1 x half> %a)
+	ret <1 x half> %res
----------------
arsenm wrote:

Overly indented

https://github.com/llvm/llvm-project/pull/137137