diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td index 9c96c080c08..c9eac7d885f 100644 --- a/lib/Target/R600/SIInstructions.td +++ b/lib/Target/R600/SIInstructions.td @@ -535,7 +535,7 @@ def IMAGE_SAMPLE_B : MIMG_Sampler_Helper <0x00000025, "IMAGE_SAMPLE_B">; //def IMAGE_SAMPLE_LZ : MIMG_NoPattern_ <"IMAGE_SAMPLE_LZ", 0x00000027>; def IMAGE_SAMPLE_C : MIMG_Sampler_Helper <0x00000028, "IMAGE_SAMPLE_C">; //def IMAGE_SAMPLE_C_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CL", 0x00000029>; -//def IMAGE_SAMPLE_C_D : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D", 0x0000002a>; +def IMAGE_SAMPLE_C_D : MIMG_Sampler_Helper <0x0000002a, "IMAGE_SAMPLE_C_D">; //def IMAGE_SAMPLE_C_D_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_CL", 0x0000002b>; def IMAGE_SAMPLE_C_L : MIMG_Sampler_Helper <0x0000002c, "IMAGE_SAMPLE_C_L">; def IMAGE_SAMPLE_C_B : MIMG_Sampler_Helper <0x0000002d, "IMAGE_SAMPLE_C_B">; @@ -1296,6 +1296,11 @@ multiclass SamplePatterns { def : SampleArrayPattern ; def : SampleShadowPattern ; def : SampleShadowArrayPattern ; + + def : SamplePattern ; + def : SampleArrayPattern ; + def : SampleShadowPattern ; + def : SampleShadowArrayPattern ; } defm : SamplePatterns; diff --git a/lib/Target/R600/SIIntrinsics.td b/lib/Target/R600/SIIntrinsics.td index 224cd2f608c..d2643e086b4 100644 --- a/lib/Target/R600/SIIntrinsics.td +++ b/lib/Target/R600/SIIntrinsics.td @@ -23,6 +23,7 @@ let TargetPrefix = "SI", isTarget = 1 in { def int_SI_sample : Sample; def int_SI_sampleb : Sample; + def int_SI_sampled : Sample; def int_SI_samplel : Sample; def int_SI_imageload : Intrinsic <[llvm_v4i32_ty], [llvm_anyvector_ty, llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>; diff --git a/test/CodeGen/R600/llvm.SI.sampled.ll b/test/CodeGen/R600/llvm.SI.sampled.ll new file mode 100644 index 00000000000..71b8ef5beaa --- /dev/null +++ b/test/CodeGen/R600/llvm.SI.sampled.ll @@ -0,0 +1,140 @@ +;RUN: llc < %s -march=r600 -mcpu=verde | FileCheck %s + +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+_VGPR[0-9]+_VGPR[0-9]+_VGPR[0-9]+}}, 15 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+_VGPR[0-9]+}}, 3 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+}}, 2 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+}}, 1 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+}}, 4 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+}}, 8 +;CHECK: IMAGE_SAMPLE_C_D {{VGPR[0-9]+_VGPR[0-9]+}}, 5 +;CHECK: IMAGE_SAMPLE_C_D {{VGPR[0-9]+_VGPR[0-9]+}}, 9 +;CHECK: IMAGE_SAMPLE_C_D {{VGPR[0-9]+_VGPR[0-9]+}}, 6 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+_VGPR[0-9]+}}, 10 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+_VGPR[0-9]+}}, 12 +;CHECK: IMAGE_SAMPLE_C_D {{VGPR[0-9]+_VGPR[0-9]+_VGPR[0-9]+}}, 7 +;CHECK: IMAGE_SAMPLE_C_D {{VGPR[0-9]+_VGPR[0-9]+_VGPR[0-9]+}}, 11 +;CHECK: IMAGE_SAMPLE_C_D {{VGPR[0-9]+_VGPR[0-9]+_VGPR[0-9]+}}, 13 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+_VGPR[0-9]+_VGPR[0-9]+}}, 14 +;CHECK: IMAGE_SAMPLE_D {{VGPR[0-9]+}}, 8 + +define void @test(i32 %a1, i32 %a2, i32 %a3, i32 %a4) { + %v1 = insertelement <4 x i32> undef, i32 %a1, i32 0 + %v2 = insertelement <4 x i32> undef, i32 %a1, i32 1 + %v3 = insertelement <4 x i32> undef, i32 %a1, i32 2 + %v4 = insertelement <4 x i32> undef, i32 %a1, i32 3 + %v5 = insertelement <4 x i32> undef, i32 %a2, i32 0 + %v6 = insertelement <4 x i32> undef, i32 %a2, i32 1 + %v7 = insertelement <4 x i32> undef, i32 %a2, i32 2 + %v8 = insertelement <4 x i32> undef, i32 %a2, i32 3 + %v9 = insertelement <4 x i32> undef, i32 %a3, i32 0 + %v10 = insertelement <4 x i32> undef, i32 %a3, i32 1 + %v11 = insertelement <4 x i32> undef, i32 %a3, i32 2 + %v12 = insertelement <4 x i32> undef, i32 %a3, i32 3 + %v13 = insertelement <4 x i32> undef, i32 %a4, i32 0 + %v14 = insertelement <4 x i32> undef, i32 %a4, i32 1 + %v15 = insertelement <4 x i32> undef, i32 %a4, i32 2 + %v16 = insertelement <4 x i32> undef, i32 %a4, i32 3 + %res1 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v1, + <8 x i32> undef, <4 x i32> undef, i32 1) + %res2 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v2, + <8 x i32> undef, <4 x i32> undef, i32 2) + %res3 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v3, + <8 x i32> undef, <4 x i32> undef, i32 3) + %res4 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v4, + <8 x i32> undef, <4 x i32> undef, i32 4) + %res5 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v5, + <8 x i32> undef, <4 x i32> undef, i32 5) + %res6 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v6, + <8 x i32> undef, <4 x i32> undef, i32 6) + %res7 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v7, + <8 x i32> undef, <4 x i32> undef, i32 7) + %res8 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v8, + <8 x i32> undef, <4 x i32> undef, i32 8) + %res9 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v9, + <8 x i32> undef, <4 x i32> undef, i32 9) + %res10 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v10, + <8 x i32> undef, <4 x i32> undef, i32 10) + %res11 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v11, + <8 x i32> undef, <4 x i32> undef, i32 11) + %res12 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v12, + <8 x i32> undef, <4 x i32> undef, i32 12) + %res13 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v13, + <8 x i32> undef, <4 x i32> undef, i32 13) + %res14 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v14, + <8 x i32> undef, <4 x i32> undef, i32 14) + %res15 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v15, + <8 x i32> undef, <4 x i32> undef, i32 15) + %res16 = call <4 x float> @llvm.SI.sampled.(<4 x i32> %v16, + <8 x i32> undef, <4 x i32> undef, i32 16) + %e1 = extractelement <4 x float> %res1, i32 0 + %e2 = extractelement <4 x float> %res2, i32 1 + %e3 = extractelement <4 x float> %res3, i32 2 + %e4 = extractelement <4 x float> %res4, i32 3 + %t0 = extractelement <4 x float> %res5, i32 0 + %t1 = extractelement <4 x float> %res5, i32 1 + %e5 = fadd float %t0, %t1 + %t2 = extractelement <4 x float> %res6, i32 0 + %t3 = extractelement <4 x float> %res6, i32 2 + %e6 = fadd float %t2, %t3 + %t4 = extractelement <4 x float> %res7, i32 0 + %t5 = extractelement <4 x float> %res7, i32 3 + %e7 = fadd float %t4, %t5 + %t6 = extractelement <4 x float> %res8, i32 1 + %t7 = extractelement <4 x float> %res8, i32 2 + %e8 = fadd float %t6, %t7 + %t8 = extractelement <4 x float> %res9, i32 1 + %t9 = extractelement <4 x float> %res9, i32 3 + %e9 = fadd float %t8, %t9 + %t10 = extractelement <4 x float> %res10, i32 2 + %t11 = extractelement <4 x float> %res10, i32 3 + %e10 = fadd float %t10, %t11 + %t12 = extractelement <4 x float> %res11, i32 0 + %t13 = extractelement <4 x float> %res11, i32 1 + %t14 = extractelement <4 x float> %res11, i32 2 + %t15 = fadd float %t12, %t13 + %e11 = fadd float %t14, %t15 + %t16 = extractelement <4 x float> %res12, i32 0 + %t17 = extractelement <4 x float> %res12, i32 1 + %t18 = extractelement <4 x float> %res12, i32 3 + %t19 = fadd float %t16, %t17 + %e12 = fadd float %t18, %t19 + %t20 = extractelement <4 x float> %res13, i32 0 + %t21 = extractelement <4 x float> %res13, i32 2 + %t22 = extractelement <4 x float> %res13, i32 3 + %t23 = fadd float %t20, %t21 + %e13 = fadd float %t22, %t23 + %t24 = extractelement <4 x float> %res14, i32 1 + %t25 = extractelement <4 x float> %res14, i32 2 + %t26 = extractelement <4 x float> %res14, i32 3 + %t27 = fadd float %t24, %t25 + %e14 = fadd float %t26, %t27 + %t28 = extractelement <4 x float> %res15, i32 0 + %t29 = extractelement <4 x float> %res15, i32 1 + %t30 = extractelement <4 x float> %res15, i32 2 + %t31 = extractelement <4 x float> %res15, i32 3 + %t32 = fadd float %t28, %t29 + %t33 = fadd float %t30, %t31 + %e15 = fadd float %t32, %t33 + %e16 = extractelement <4 x float> %res16, i32 3 + %s1 = fadd float %e1, %e2 + %s2 = fadd float %s1, %e3 + %s3 = fadd float %s2, %e4 + %s4 = fadd float %s3, %e5 + %s5 = fadd float %s4, %e6 + %s6 = fadd float %s5, %e7 + %s7 = fadd float %s6, %e8 + %s8 = fadd float %s7, %e9 + %s9 = fadd float %s8, %e10 + %s10 = fadd float %s9, %e11 + %s11 = fadd float %s10, %e12 + %s12 = fadd float %s11, %e13 + %s13 = fadd float %s12, %e14 + %s14 = fadd float %s13, %e15 + %s15 = fadd float %s14, %e16 + call void @llvm.SI.export(i32 15, i32 0, i32 1, i32 12, i32 0, float %s15, float %s15, float %s15, float %s15) + ret void +} + +declare <4 x float> @llvm.SI.sampled.(<4 x i32>, <8 x i32>, <4 x i32>, i32) readnone + +declare void @llvm.SI.export(i32, i32, i32, i32, i32, float, float, float, float)