vkd3d-shader/hlsl: Implement the InterlockedMin() intrinsic.

Approved-by: Henri Verbeet (@hverbeet)
Approved-by: Elizabeth Figura (@zfigura)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/1330

libs/vkd3d-shader/hlsl.c

@@ -3511,6 +3511,7 @@ static void dump_ir_interlocked(struct vkd3d_string_buffer *buffer, const struct
         [HLSL_INTERLOCKED_CMP_EXCH] = "cmp_exch",
         [HLSL_INTERLOCKED_EXCH] = "exch",
         [HLSL_INTERLOCKED_MAX] = "max",
+        [HLSL_INTERLOCKED_MIN] = "min",
     };
 
     VKD3D_ASSERT(interlocked->op < ARRAY_SIZE(op_names));

libs/vkd3d-shader/hlsl.h

@@ -958,6 +958,7 @@ enum hlsl_interlocked_op
     HLSL_INTERLOCKED_CMP_EXCH,
     HLSL_INTERLOCKED_EXCH,
     HLSL_INTERLOCKED_MAX,
+    HLSL_INTERLOCKED_MIN,
 };
 
 /* Represents an interlocked operation.

libs/vkd3d-shader/hlsl.y

@@ -5429,7 +5429,7 @@ static bool intrinsic_interlocked(struct hlsl_ctx *ctx, enum hlsl_interlocked_op
 
     /* Interlocked*() functions always take uint for the value parameters,
      * except for InterlockedMax()/InterlockedMin(). */
-    if (op == HLSL_INTERLOCKED_MAX)
+    if (op == HLSL_INTERLOCKED_MAX || op == HLSL_INTERLOCKED_MIN)
     {
         enum hlsl_base_type val_base_type = val->data_type->e.numeric.type;
 
@@ -5536,6 +5536,12 @@ static bool intrinsic_InterlockedMax(struct hlsl_ctx *ctx,
     return intrinsic_interlocked(ctx, HLSL_INTERLOCKED_MAX, params, loc, "InterlockedMax");
 }
 
+static bool intrinsic_InterlockedMin(struct hlsl_ctx *ctx,
+        const struct parse_initializer *params, const struct vkd3d_shader_location *loc)
+{
+    return intrinsic_interlocked(ctx, HLSL_INTERLOCKED_MIN, params, loc, "InterlockedMin");
+}
+
 static const struct intrinsic_function
 {
     const char *name;
@@ -5555,6 +5561,7 @@ intrinsic_functions[] =
     {"InterlockedCompareStore",             3, true,  intrinsic_InterlockedCompareStore},
     {"InterlockedExchange",                 3, true,  intrinsic_InterlockedExchange},
     {"InterlockedMax",                     -1, true,  intrinsic_InterlockedMax},
+    {"InterlockedMin",                     -1, true,  intrinsic_InterlockedMin},
     {"abs",                                 1, true,  intrinsic_abs},
     {"acos",                                1, true,  intrinsic_acos},
     {"all",                                 1, true,  intrinsic_all},

libs/vkd3d-shader/hlsl_codegen.c

@@ -10004,6 +10004,7 @@ static bool sm4_generate_vsir_instr_interlocked(struct hlsl_ctx *ctx,
         [HLSL_INTERLOCKED_AND] = VKD3DSIH_ATOMIC_AND,
         [HLSL_INTERLOCKED_CMP_EXCH] = VKD3DSIH_ATOMIC_CMP_STORE,
         [HLSL_INTERLOCKED_MAX] = VKD3DSIH_ATOMIC_UMAX,
+        [HLSL_INTERLOCKED_MIN] = VKD3DSIH_ATOMIC_UMIN,
     };
 
     static const enum vkd3d_shader_opcode imm_opcodes[] =
@@ -10013,6 +10014,7 @@ static bool sm4_generate_vsir_instr_interlocked(struct hlsl_ctx *ctx,
         [HLSL_INTERLOCKED_CMP_EXCH] = VKD3DSIH_IMM_ATOMIC_CMP_EXCH,
         [HLSL_INTERLOCKED_EXCH] = VKD3DSIH_IMM_ATOMIC_EXCH,
         [HLSL_INTERLOCKED_MAX] = VKD3DSIH_IMM_ATOMIC_UMAX,
+        [HLSL_INTERLOCKED_MIN] = VKD3DSIH_IMM_ATOMIC_UMIN,
     };
 
     struct hlsl_ir_node *cmp_value = interlocked->cmp_value.node, *value = interlocked->value.node;
@@ -10029,8 +10031,12 @@ static bool sm4_generate_vsir_instr_interlocked(struct hlsl_ctx *ctx,
     {
         if (opcode == VKD3DSIH_ATOMIC_UMAX)
             opcode = VKD3DSIH_ATOMIC_IMAX;
+        else if (opcode == VKD3DSIH_ATOMIC_UMIN)
+            opcode = VKD3DSIH_ATOMIC_IMIN;
         else if (opcode == VKD3DSIH_IMM_ATOMIC_UMAX)
             opcode = VKD3DSIH_IMM_ATOMIC_IMAX;
+        else if (opcode == VKD3DSIH_IMM_ATOMIC_UMIN)
+            opcode = VKD3DSIH_IMM_ATOMIC_IMIN;
     }
 
     if (!(ins = generate_vsir_add_program_instruction(ctx, program, &instr->loc, opcode,

libs/vkd3d-shader/tpf.c

@@ -4030,7 +4030,9 @@ static void tpf_handle_instruction(struct tpf_compiler *tpf, const struct vkd3d_
         case VKD3DSIH_ATOMIC_CMP_STORE:
         case VKD3DSIH_ATOMIC_IADD:
         case VKD3DSIH_ATOMIC_IMAX:
+        case VKD3DSIH_ATOMIC_IMIN:
         case VKD3DSIH_ATOMIC_UMAX:
+        case VKD3DSIH_ATOMIC_UMIN:
         case VKD3DSIH_AND:
         case VKD3DSIH_BREAK:
         case VKD3DSIH_CASE:
@@ -4078,7 +4080,9 @@ static void tpf_handle_instruction(struct tpf_compiler *tpf, const struct vkd3d_
         case VKD3DSIH_IMM_ATOMIC_EXCH:
         case VKD3DSIH_IMM_ATOMIC_IADD:
         case VKD3DSIH_IMM_ATOMIC_IMAX:
+        case VKD3DSIH_IMM_ATOMIC_IMIN:
         case VKD3DSIH_IMM_ATOMIC_UMAX:
+        case VKD3DSIH_IMM_ATOMIC_UMIN:
         case VKD3DSIH_IMUL:
         case VKD3DSIH_INE:
         case VKD3DSIH_INEG:

tests/hlsl/uav-atomics.shader_test

@@ -66,7 +66,7 @@ size (buffer, 2)
 
 -3 1
 
-[compute shader fail(sm<5) todo(sm>=5)]
+[compute shader fail(sm<5)]
 RWBuffer<int> u : register(u2);
 
 uniform int4 i;
@@ -80,12 +80,12 @@ void main()
 
 [test]
 uniform 0 int4 1 -3 0 0
-todo(sm<6) dispatch 1 1 1
+todo(glsl | msl) dispatch 1 1 1
 probe uav 2 (0) ri (1)
 probe uav 2 (1) ri (-3)
 
 uniform 0 int4 -3 1 0 0
-todo(sm<6) dispatch 1 1 1
+todo(glsl | msl) dispatch 1 1 1
 probe uav 2 (0) ri (1)
 probe uav 2 (1) ri (-3)
 
@@ -134,7 +134,7 @@ probe uav 1 (9) rui (3)
 probe uav 1 (10) rui (1)
 
 
-[compute shader fail(sm<5) todo(sm>=5)]
+[compute shader fail(sm<5)]
 RWTexture2D<int> u : register(u2);
 
 uniform int4 i;
@@ -148,12 +148,13 @@ void main()
 
 [test]
 uniform 0 int4 1 -3 0 0
-todo(sm<6) dispatch 1 1 1
+% SPIR-V compilation currently fails because of the mismatched resource type for u2.
+todo(mvk | vulkan | opengl | msl) dispatch 1 1 1
 probe uav 2 (0) ri (1)
 probe uav 2 (1) ri (-3)
 
 uniform 0 int4 -3 1 0 0
-todo(sm<6) dispatch 1 1 1
+todo(mvk | vulkan | opengl | msl) dispatch 1 1 1
 probe uav 2 (0) ri (1)
 probe uav 2 (1) ri (-3)
 
@@ -173,7 +174,7 @@ size (buffer, 3)
 % The value fields of InterlockedMax/Min have the same type as the underlying scalar type of dst.
 % However, floating point numbers are always converted to signed integers.
 
-[compute shader fail(sm<5) todo(sm>=5)]
+[compute shader fail(sm<5)]
 RWBuffer<uint> u : register(u1);
 RWBuffer<int> s : register(u2);
 
@@ -195,7 +196,7 @@ void main()
 [test]
 uniform 0 uint  0xffffffff
 uniform 1 float 0x80000000
-todo(sm<6) dispatch 1 1 1
+todo(glsl | msl) dispatch 1 1 1
 probe uav 1 (0) rui (0xffffffff)
 probe uav 1 (1) rui (1)
 probe uav 1 (2) rui (0x7fffffff)