16bit for asm inline reg

Author: broxigarchen

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

@@ -16062,7 +16062,8 @@ SITargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI_,
     case 'v':
       switch (BitWidth) {
       case 16:
-        RC = &AMDGPU::VGPR_32RegClass;
+        RC = Subtarget->useRealTrue16Insts() ? &AMDGPU::VGPR_16RegClass
+                                             : &AMDGPU::VGPR_32RegClass;
         break;
       default:
         RC = TRI->getVGPRClassForBitWidth(BitWidth);

llvm/test/CodeGen/AMDGPU/inlineasm-16-fake16.ll

@@ -0,0 +1,48 @@
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 -verify-machineinstrs < %s 2>&1 | FileCheck -enable-var-scope -check-prefixes=GFX11 %s
+
+; GFX11-LABEL: {{^}}s_input_output_i16:
+; GFX11: s_mov_b32 s[[REG:[0-9]+]], -1
+; GFX11: ; use s[[REG]]
+define amdgpu_kernel void @s_input_output_i16() #0 {
+  %v = tail call i16 asm sideeffect "s_mov_b32 $0, -1", "=s"()
+  tail call void asm sideeffect "; use $0", "s"(i16 %v) #0
+  ret void
+}
+
+; GFX11-LABEL: {{^}}s_input_output_f16:
+; GFX11: s_mov_b32 s[[REG:[0-9]+]], -1
+; GFX11: ; use s[[REG]]
+define amdgpu_kernel void @s_input_output_f16() #0 {
+  %v = tail call half asm sideeffect "s_mov_b32 $0, -1", "=s"() #0
+  tail call void asm sideeffect "; use $0", "s"(half %v)
+  ret void
+}
+
+; GFX11-LABEL: {{^}}v_input_output_f16:
+; GFX11: v_mov_b32 v[[REG:[0-9]+]], -1
+; GFX11: ; use v[[REG]]
+define amdgpu_kernel void @v_input_output_f16() #0 {
+  %v = tail call half asm sideeffect "v_mov_b32 $0, -1", "=v"() #0
+  tail call void asm sideeffect "; use $0", "v"(half %v)
+  ret void
+}
+
+; GFX11-LABEL: {{^}}v_input_output_i16:
+; GFX11: v_mov_b32 v[[REG:[0-9]+]], -1
+; GFX11: ; use v[[REG]]
+define amdgpu_kernel void @v_input_output_i16() #0 {
+  %v = tail call i16 asm sideeffect "v_mov_b32 $0, -1", "=v"() #0
+  tail call void asm sideeffect "; use $0", "v"(i16 %v)
+  ret void
+}
+
+; GFX11-LABEL: {{^}}i16_imm_input_phys_vgpr:
+; GFX11: v_mov_b32_e32 v0, 0xffff
+; GFX11: ; use v0
+define amdgpu_kernel void @i16_imm_input_phys_vgpr() {
+entry:
+  call void asm sideeffect "; use $0 ", "{v0}"(i16 65535)
+  ret void
+}
+
+attributes #0 = { nounwind }

llvm/test/CodeGen/AMDGPU/inlineasm-16-true16.ll

@@ -0,0 +1,57 @@
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 -verify-machineinstrs < %s 2>&1 | FileCheck -enable-var-scope -check-prefixes=GFX11 %s
+
+; GFX11-LABEL: {{^}}s_input_output_i16:
+; GFX11: s_mov_b32 s[[REG:[0-9]+]], -1
+; GFX11: ; use s[[REG]]
+define amdgpu_kernel void @s_input_output_i16() #0 {
+  %v = tail call i16 asm sideeffect "s_mov_b32 $0, -1", "=s"()
+  tail call void asm sideeffect "; use $0", "s"(i16 %v) #0
+  ret void
+}
+
+; GFX11-LABEL: {{^}}s_input_output_f16:
+; GFX11: s_mov_b32 s[[REG:[0-9]+]], -1
+; GFX11: ; use s[[REG]]
+define amdgpu_kernel void @s_input_output_f16() #0 {
+  %v = tail call half asm sideeffect "s_mov_b32 $0, -1", "=s"() #0
+  tail call void asm sideeffect "; use $0", "s"(half %v)
+  ret void
+}
+
+; GFX11-LABEL: {{^}}v_input_output_f16:
+; GFX11: v_mov_b16 v[[REG:[0-9]+.(l|h)]], -1
+; GFX11: ; use v[[REG]]
+define amdgpu_kernel void @v_input_output_f16() #0 {
+  %v = tail call half asm sideeffect "v_mov_b16 $0, -1", "=v"() #0
+  tail call void asm sideeffect "; use $0", "v"(half %v)
+  ret void
+}
+
+; GFX11-LABEL: {{^}}v_input_output_i16:
+; GFX11: v_mov_b16 v[[REG:[0-9]+.(l|h)]], -1
+; GFX11: ; use v[[REG]]
+define amdgpu_kernel void @v_input_output_i16() #0 {
+  %v = tail call i16 asm sideeffect "v_mov_b16 $0, -1", "=v"() #0
+  tail call void asm sideeffect "; use $0", "v"(i16 %v)
+  ret void
+}
+
+; GFX11-LABEL: {{^}}i16_imm_input_phys_vgpr_lo:
+; GFX11: v_mov_b16_e32 v0.l, -1
+; GFX11: ; use v0.l
+define amdgpu_kernel void @i16_imm_input_phys_vgpr_lo() {
+entry:
+  call void asm sideeffect "; use $0 ", "{v0.l}"(i16 65535)
+  ret void
+}
+
+; GFX11-LABEL: {{^}}i16_imm_input_phys_vgpr_hi:
+; GFX11: v_mov_b16_e32 v0.h, -1
+; GFX11: ; use v0.h
+define amdgpu_kernel void @i16_imm_input_phys_vgpr_hi() {
+entry:
+  call void asm sideeffect "; use $0 ", "{v0.h}"(i16 65535)
+  ret void
+}
+
+attributes #0 = { nounwind }