CUDA/C++ backend improvements (#1198)

* WIP with vector float test.

* vector-float test working.

* Fixed remaing tests broken with init changes.

* Improve 64bit-type-support.md

* Disable tests broken on CI system for Dx.

* WIP: Make type available for comparison.

* Moved type conversion into TypeTextUtil.

* Add text/type conversions from DownstreamCompiler to TypeTextUtil.

* Allow compaison taking into account type.

* Removed quantize in vector-float.slang test.

8 files changed, 138 insertions, 5 deletions

diff --git a/tests/bugs/dxbc-double-problem.slang.expected.txt b/tests/bugs/dxbc-double-problem.slang.expected.txt
index 2cffc0e7c..d560391b9 100644
--- a/tests/bugs/dxbc-double-problem.slang.expected.txt
+++ b/tests/bugs/dxbc-double-problem.slang.expected.txt
@@ -1,3 +1,4 @@
+type: double
 0.674277
 3.141593
 0.017453
diff --git a/tests/hlsl-intrinsic/scalar-double-d3d-intrinsic.slang b/tests/hlsl-intrinsic/scalar-double-d3d-intrinsic.slang
index 02bffb7b1..600dfa88d 100644
--- a/tests/hlsl-intrinsic/scalar-double-d3d-intrinsic.slang
+++ b/tests/hlsl-intrinsic/scalar-double-d3d-intrinsic.slang
@@ -1,8 +1,7 @@
-// It doesn't look like fxc, dxc, vk support double versions of many of the intrinsics, so they are disabled here.
-
 //TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute -output-using-type
+// On dx11 this causes a crash on CI (not currently not repeatable on test systems)
 //DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -output-using-type
-// Fxc doesn't have double intrinsics available
+// This is disabled because Dx12/DXBC doing double writes to a structured buffer can fail
 //DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -output-using-type
 //TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -profile sm_6_0 -use-dxil -output-using-type
 //TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -output-using-type
diff --git a/tests/hlsl-intrinsic/scalar-double-d3d-intrinsic.slang.expected.txt b/tests/hlsl-intrinsic/scalar-double-d3d-intrinsic.slang.expected.txt
index 60826a847..44f39ae21 100644
--- a/tests/hlsl-intrinsic/scalar-double-d3d-intrinsic.slang.expected.txt
+++ b/tests/hlsl-intrinsic/scalar-double-d3d-intrinsic.slang.expected.txt
@@ -1,3 +1,4 @@
+type: double
 2.850000
 2.550000
 3.216667
diff --git a/tests/hlsl-intrinsic/scalar-double-vk-intrinsic.slang.expected.txt b/tests/hlsl-intrinsic/scalar-double-vk-intrinsic.slang.expected.txt
index 6fd20cd00..b1b17082a 100644
--- a/tests/hlsl-intrinsic/scalar-double-vk-intrinsic.slang.expected.txt
+++ b/tests/hlsl-intrinsic/scalar-double-vk-intrinsic.slang.expected.txt
@@ -1,3 +1,4 @@
+type: double
 -105.150000
 -46.277573
 13.888270
diff --git a/tests/hlsl-intrinsic/scalar-int64-literal-problem.slang b/tests/hlsl-intrinsic/scalar-int64-literal-problem.slang
index 71f058dc1..c315e84cd 100644
--- a/tests/hlsl-intrinsic/scalar-int64-literal-problem.slang
+++ b/tests/hlsl-intrinsic/scalar-int64-literal-problem.slang
@@ -1,12 +1,15 @@
+// Disable the test in general, exists to check the output when the fix is in.
+
 //TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
 // No support for int64_t on dx11 (no sm 6.0)
 //DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
 // No support with Dx12 with dxbc. Needs SM6.0 + dxil
 //DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -profile cs_6_0 -dx12 -use-dxil
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -profile cs_6_0 -dx12 -use-dxil
 // GLSL notices the narrowing of a 64 bit literal into 32 bits and for it this is an error.
 //DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
-//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/hlsl-intrinsic/scalar-int64-literal-problem.slang.expected.txt b/tests/hlsl-intrinsic/scalar-int64-literal-problem.slang.expected.txt
index e69de29bb..6c2c9dabe 100644
--- a/tests/hlsl-intrinsic/scalar-int64-literal-problem.slang.expected.txt
+++ b/tests/hlsl-intrinsic/scalar-int64-literal-problem.slang.expected.txt
@@ -0,0 +1,8 @@
+7FFFFFFF
+FFFFFFFF
+0
+FFFFFFFD
+80000000
+FFFFFFFB
+0
+FFFFFFF9
diff --git a/tests/hlsl-intrinsic/vector-float.slang b/tests/hlsl-intrinsic/vector-float.slang
new file mode 100644
index 000000000..2ca98de6c
--- /dev/null
+++ b/tests/hlsl-intrinsic/vector-float.slang
@@ -0,0 +1,103 @@
+//TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute -output-using-type
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -output-using-type
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -output-using-type
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -output-using-type
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-cuda -compute -output-using-type
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
+RWStructuredBuffer<float4> outputBuffer;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    int idx = int(dispatchThreadID.x);
+
+    float vf = idx * (1.0f / (4.0f));
+     
+    float3 f = float3(0.1f, vf, vf + 0.2f);
+    
+    // Vector specific
+    float3 nv = normalize(f);
+    
+    
+    // Operate over all values
+    float3 ft = {};
+
+    // fmod
+    ft += float3(int3(((f % 0.11f) * 100) + 0.5));
+    
+    ft += sin(f);
+    ft += cos(f);
+    ft += tan(f);
+    
+    ft += asin(f);
+    ft += acos(f);
+    ft += atan(f);
+    
+    ft += atan2(f, 2.0);
+
+#if 0
+    {
+        // Disabled because not supported on VK (glsl) in vector form
+        float3 sf, cf;
+        sincos(f, sf, cf);
+        
+        ft += sf;
+        ft += cf;
+    }
+#endif     
+     
+#if 0     
+    // Disabled because not supported on VK (glsl) in vector form
+    ft += rcp(1.0 + f);
+#endif
+
+    ft += sign(f - 0.5);
+    
+    ft += saturate(f * 4 - 2.0);
+    
+    ft += sqrt(f);
+    ft += rsqrt(1.0f + f);
+    
+    ft += exp2(f);
+    ft += exp(f);
+     
+    ft += frac(f * 3);
+    ft += ceil(f * 5 - 3);
+    
+    ft += floor(f * 10 - 7.01);
+
+    ft += trunc(f * 7);
+   
+    ft += log(f + 10.0);
+    ft += log2(f * 3 + 2);
+
+
+    {
+        float v[] = { 1, 10, 100, 1000 };
+        ft += int3(log10(float3(v[idx] + vf) + 0.5f));
+    }
+
+       
+    ft += abs(f * 4 - 2.0f);
+    
+    ft += min(0.5, f);
+    ft += max(f, 0.75);
+
+    ft += pow(0.5, f);
+
+    ft += smoothstep(0.2, 0.7, f);
+    ft += lerp(-100, 100, f);
+
+    ft += clamp(f, 0.1, 0.3);
+
+    ft += step(f, 0.5);
+
+    int3 vi = asint(f - f) + idx;
+    
+    ft += float3(vi);   
+
+    float4 result = float4(ft + nv, 0) + float4(vf);
+
+    outputBuffer[idx] = result;
+}
\ No newline at end of file
diff --git a/tests/hlsl-intrinsic/vector-float.slang.expected.txt b/tests/hlsl-intrinsic/vector-float.slang.expected.txt
new file mode 100644
index 000000000..c2a569574
--- /dev/null
+++ b/tests/hlsl-intrinsic/vector-float.slang.expected.txt
@@ -0,0 +1,17 @@
+type: float
+-64.893356
+-98.276619
+-42.604630
+0.000000
+-62.899876
+-34.338215
+9.949981
+0.250000
+-60.725098
+28.410460
+73.591156
+0.500000
+-58.508228
+93.563255
+138.897034
+0.750000